Skip to content

P
PaddleHub

PaddlePaddle / PaddleHub
大约 2 年 前同步成功

通知 285
Star 12117
Fork 2091
P
PaddleHub
提交 d6b6c6ff 编写于 作者: W wuzewu
浏览文件
操作

Fix travis-ci issue.

上级 b100dc5d
显示空白变更内容
内联 并排
Showing with 1919 addition and 2895 deletion
.travis.yml
浏览文件 @ d6b6c6ff
......@@ -27,6 +27,7 @@ install:
else
pip install --upgrade paddlepaddle;
pip install -r requirements.txt;
pip install yapf==0.26.0
fi
notifications:
......
demo/audio_classification/predict.py
浏览文件 @ d6b6c6ff
......@@ -26,16 +26,15 @@ parser.add_argument("--wav", type=str, required=True, help="Audio file to infer.
parser.add_argument("--sr", type=int, default=44100, help="Sample rate of inference audio.")
parser.add_argument("--model_type", type=str, default='panns_cnn14', help="Select model to to inference.")
parser.add_argument("--topk", type=int, default=1, help="Show top k results of prediction labels.")
parser.add_argument("--checkpoint",
type=str,
default='./checkpoint/best_model/model.pdparams',
help="Checkpoint of model.")
parser.add_argument(
"--checkpoint", type=str, default='./checkpoint/best_model/model.pdparams', help="Checkpoint of model.")
args = parser.parse_args()
if __name__ == '__main__':
label_map = {idx: label for idx, label in enumerate(ESC50.label_list)}
model = hub.Module(name=args.model_type,
model = hub.Module(
name=args.model_type,
version='1.0.0',
task='sound-cls',
num_class=ESC50.num_class,
......
demo/audio_classification/train.py
浏览文件 @ d6b6c6ff
......@@ -22,7 +22,8 @@ from paddlehub.datasets import ESC50
parser = argparse.ArgumentParser(__doc__)
parser.add_argument("--num_epoch", type=int, default=50, help="Number of epoches for fine-tuning.")
parser.add_argument("--use_gpu",
parser.add_argument(
"--use_gpu",
type=ast.literal_eval,
default=True,
help="Whether use GPU for fine-tuning, input should be True or False")
......
demo/autoaug/README.md
浏览文件 @ d6b6c6ff
......@@ -173,8 +173,3 @@ bash search.sh
cd PaddleHub/demo/autaug/
bash train.sh
```
demo/autoaug/hub_fitter.py
浏览文件 @ d6b6c6ff
......@@ -23,6 +23,7 @@ logger = log.get_logger(level=logging.INFO)
import auto_augment
auto_augment_path = auto_augment.__file__
class HubFitterClassifer(object):
"""Trains an instance of the Model class."""
......@@ -43,8 +44,7 @@ class HubFitterClassifer(object):
# `import paddle`. Otherwise, it would not take any effect.
set_paddle_flags(
# enable GC to save memory
FLAGS_fraction_of_gpu_memory_to_use=hparams.resource_config.gpu,
)
FLAGS_fraction_of_gpu_memory_to_use=hparams.resource_config.gpu, )
import paddle
import paddlehub as hub
from paddlehub_utils.trainer import CustomTrainer
......@@ -63,23 +63,18 @@ class HubFitterClassifer(object):
paddle.disable_static(paddle.CUDAPlace(paddle.distributed.get_rank()))
train_dataset, eval_dataset = _init_loader(self.hparams)
model = hub.Module(name=hparams["task_config"]["classifier"]["model_name"], label_list=self.class_to_id_dict.keys(), load_checkpoint=None)
optimizer = paddle.optimizer.Adam(
learning_rate=0.001, parameters=model.parameters())
trainer = CustomTrainer(
model=model,
optimizer=optimizer,
checkpoint_dir='img_classification_ckpt')
model = hub.Module(
name=hparams["task_config"]["classifier"]["model_name"],
label_list=self.class_to_id_dict.keys(),
load_checkpoint=None)
optimizer = paddle.optimizer.Adam(learning_rate=0.001, parameters=model.parameters())
trainer = CustomTrainer(model=model, optimizer=optimizer, checkpoint_dir='img_classification_ckpt')
self.model = model
self.optimizer = optimizer
trainer.init_train_and_eval(
train_dataset,
epochs=100,
batch_size=32,
eval_dataset=eval_dataset,
save_interval=1)
train_dataset, epochs=100, batch_size=32, eval_dataset=eval_dataset, save_interval=1)
self.trainer = trainer
def _fit_param(self, show: bool = False) -> None:
......@@ -123,8 +118,7 @@ class HubFitterClassifer(object):
"""
self.hparams = new_hparams
self.trainer.train_loader.dataset.reset_policy(
new_hparams.search_space)
self.trainer.train_loader.dataset.reset_policy(new_hparams.search_space)
return None
def save_model(self, checkpoint_dir: str, step: Optional[str] = None) -> str:
......@@ -134,8 +128,7 @@ class HubFitterClassifer(object):
step: If provided, creates a checkpoint with the given step
number, instead of overwriting the existing checkpoints.
"""
checkpoint_path = os.path.join(checkpoint_dir,
'epoch') + '-' + str(step)
checkpoint_path = os.path.join(checkpoint_dir, 'epoch') + '-' + str(step)
logger.info('Saving model checkpoint to {}'.format(checkpoint_path))
self.trainer.save_model(os.path.join(checkpoint_path, "checkpoint"))
......@@ -145,8 +138,7 @@ class HubFitterClassifer(object):
"""Loads a checkpoint with the architecture structure stored in the name."""
ckpt_path = os.path.join(checkpoint_path, "checkpoint")
self.trainer.load_model(ckpt_path)
logger.info(
'Loaded child model checkpoint from {}'.format(checkpoint_path))
logger.info('Loaded child model checkpoint from {}'.format(checkpoint_path))
def eval_child_model(self, mode: str, pass_id: int = 0) -> dict:
"""Evaluate the child model.
......@@ -204,10 +196,7 @@ class HubFitterClassifer(object):
"""Trains the model `m` for one epoch."""
start_time = time.time()
train_acc = self.train_one_epoch(curr_epoch)
logger.info(
'Epoch:{} time(min): {}'.format(
curr_epoch,
(time.time() - start_time) / 60.0))
logger.info('Epoch:{} time(min): {}'.format(curr_epoch, (time.time() - start_time) / 60.0))
return train_acc
def _compute_final_accuracies(self, iteration: int) -> dict:
......@@ -228,8 +217,7 @@ class HubFitterClassifer(object):
self._fit_param()
train_acc = self._run_training_loop(epoch)
valid_acc = self.eval_child_model(mode="val", pass_id=epoch)
logger.info('valid acc: {}'.format(
valid_acc))
logger.info('valid acc: {}'.format(valid_acc))
all_metric = {}
all_metric.update(train_acc)
all_metric.update(valid_acc)
......
demo/autoaug/paddlehub_utils/reader.py
浏览文件 @ d6b6c6ff
......@@ -13,7 +13,6 @@
# See the License for the specific language governing permissions and
# limitations under the License.
# -*- coding: utf-8 -*-
# *******************************************************************************
#
......@@ -36,8 +35,7 @@ import paddlehub.vision.transforms as transforms
from PIL import ImageFile
from auto_augment.autoaug.transform.autoaug_transform import AutoAugTransform
ImageFile.LOAD_TRUNCATED_IMAGES = True
__imagenet_stats = {'mean': [0.485, 0.456, 0.406],
'std': [0.229, 0.224, 0.225]}
__imagenet_stats = {'mean': [0.485, 0.456, 0.406], 'std': [0.229, 0.224, 0.225]}
class PbaAugment(object):
......@@ -45,8 +43,7 @@ class PbaAugment(object):
pytorch 分类 PbaAugment transform
"""
def __init__(
self,
def __init__(self,
input_size: int = 224,
scale_size: int = 256,
normalize: Optional[list] = None,
......@@ -64,27 +61,20 @@ class PbaAugment(object):
"""
if normalize is None:
normalize = {
'mean': [
0.485, 0.456, 0.406], 'std': [
0.229, 0.224, 0.225]}
normalize = {'mean': [0.485, 0.456, 0.406], 'std': [0.229, 0.224, 0.225]}
policy = kwargs["policy"]
assert stage in ["search", "train"]
train_epochs = kwargs["hp_policy_epochs"]
self.auto_aug_transform = AutoAugTransform.create(
policy, stage=stage, train_epochs=train_epochs)
self.auto_aug_transform = AutoAugTransform.create(policy, stage=stage, train_epochs=train_epochs)
#self.auto_aug_transform = PbtAutoAugmentClassiferTransform(conf)
if pre_transform:
self.pre_transform = transforms.Resize(input_size)
self.post_transform = transforms.Compose(
transforms=[
transforms.Permute(),
transforms.Normalize(**normalize, channel_first=True)
],
channel_first = False
)
transforms=[transforms.Permute(),
transforms.Normalize(**normalize, channel_first=True)],
channel_first=False)
self.cur_epoch = 0
def set_epoch(self, indx: int) -> None:
......@@ -164,8 +154,7 @@ class PicReader(paddle.io.Dataset):
PicReader
"""
def __init__(
self,
def __init__(self,
root_path: str,
list_file: str,
meta: bool = False,
......@@ -222,8 +211,7 @@ class PicReader(paddle.io.Dataset):
img = cv2.resize(img, (scale_size, scale_size))
self.cache_img_buff[image_path] = img
except BaseException:
print("img_path:{} can not by cv2".format(
image_path).format(image_path))
print("img_path:{} can not by cv2".format(image_path).format(image_path))
pass
......@@ -265,9 +253,7 @@ class PicReader(paddle.io.Dataset):
with open(self.list_file) as f:
lines = f.read().splitlines()
print(
"PicReader:: found {} picture in `{}'".format(
len(lines), self.list_file))
print("PicReader:: found {} picture in `{}'".format(len(lines), self.list_file))
for i, line in enumerate(lines):
record = re.split(delimiter, line)
# record = line.split()
......@@ -401,20 +387,12 @@ def _read_classes(csv_file: str) -> dict:
class_name = row.strip()
# print(class_id, class_name)
except ValueError:
six.raise_from(
ValueError(
'line {}: format should be \'class_name\''.format(line)),
None)
six.raise_from(ValueError('line {}: format should be \'class_name\''.format(line)), None)
class_id = _parse(
line,
int,
'line {}: malformed class ID: {{}}'.format(line))
class_id = _parse(line, int, 'line {}: malformed class ID: {{}}'.format(line))
if class_name in result:
raise ValueError(
'line {}: duplicate class name: \'{}\''.format(
line, class_name))
raise ValueError('line {}: duplicate class name: \'{}\''.format(line, class_name))
result[class_name] = class_id
return result
......@@ -439,10 +417,7 @@ def _init_loader(hparams: dict, TrainTransform=None) -> tuple:
epochs = hparams.task_config.classifier.epochs
no_cache_img = hparams.task_config.classifier.get("no_cache_img", False)
normalize = {
'mean': [
0.485, 0.456, 0.406], 'std': [
0.229, 0.224, 0.225]}
normalize = {'mean': [0.485, 0.456, 0.406], 'std': [0.229, 0.224, 0.225]}
if TrainTransform is None:
TrainTransform = PbaAugment(
......@@ -453,10 +428,7 @@ def _init_loader(hparams: dict, TrainTransform=None) -> tuple:
hp_policy_epochs=epochs,
)
delimiter = hparams.data_config.delimiter
kwargs = dict(
conf=hparams,
delimiter=delimiter
)
kwargs = dict(conf=hparams, delimiter=delimiter)
if hparams.task_config.classifier.use_class_map:
class_to_id_dict = _read_classes(label_list=hparams.data_config.label_list)
......@@ -475,13 +447,11 @@ def _init_loader(hparams: dict, TrainTransform=None) -> tuple:
list_file=val_list,
transform=transforms.Compose(
transforms=[
transforms.Resize(
(224,
224)),
transforms.Resize((224, 224)),
transforms.Permute(),
transforms.Normalize(
**normalize, channel_first=True)],
channel_first = False),
transforms.Normalize(**normalize, channel_first=True)
],
channel_first=False),
class_to_id_dict=class_to_id_dict,
cache_img=not no_cache_img,
**kwargs)
......
demo/autoaug/paddlehub_utils/trainer.py
浏览文件 @ d6b6c6ff
......@@ -18,6 +18,7 @@ from paddle.distributed import ParallelEnv
from paddlehub.utils.log import logger
from paddlehub.utils.utils import Timer
class CustomTrainer(Trainer):
def __init__(self, **kwargs) -> None:
super(CustomTrainer, self).__init__(**kwargs)
......@@ -33,10 +34,7 @@ class CustomTrainer(Trainer):
self.batch_sampler, self.train_loader = self.init_train(train_dataset, batch_size, num_workers)
self.eval_loader = self.init_evaluate(eval_dataset, batch_size, num_workers)
def init_train(self,
train_dataset: paddle.io.Dataset,
batch_size: int = 1,
num_workers: int = 0) -> tuple:
def init_train(self, train_dataset: paddle.io.Dataset, batch_size: int = 1, num_workers: int = 0) -> tuple:
use_gpu = True
place = paddle.CUDAPlace(ParallelEnv().dev_id) if use_gpu else paddle.CPUPlace()
paddle.disable_static(place)
......@@ -47,7 +45,8 @@ class CustomTrainer(Trainer):
train_dataset, batch_sampler=batch_sampler, places=place, num_workers=num_workers, return_list=True)
return batch_sampler, loader
def train_one_epoch(self, loader: paddle.io.DataLoader, timer: Timer, current_epoch: int, epochs: int, log_interval: int, steps_per_epoch: int) -> None:
def train_one_epoch(self, loader: paddle.io.DataLoader, timer: Timer, current_epoch: int, epochs: int,
log_interval: int, steps_per_epoch: int) -> None:
avg_loss = 0
avg_metrics = defaultdict(int)
self.model.train()
......@@ -70,15 +69,13 @@ class CustomTrainer(Trainer):
if self.use_vdl:
self.log_writer.add_scalar(tag='TRAIN/loss', step=timer.current_step, value=avg_loss)
print_msg = 'Epoch={}/{}, Step={}/{}'.format(current_epoch, epochs, batch_idx + 1,
steps_per_epoch)
print_msg = 'Epoch={}/{}, Step={}/{}'.format(current_epoch, epochs, batch_idx + 1, steps_per_epoch)
print_msg += ' loss={:.4f}'.format(avg_loss)
for metric, value in avg_metrics.items():
value /= log_interval
if self.use_vdl:
self.log_writer.add_scalar(
tag='TRAIN/{}'.format(metric), step=timer.current_step, value=value)
self.log_writer.add_scalar(tag='TRAIN/{}'.format(metric), step=timer.current_step, value=value)
print_msg += ' {}={:.4f}'.format(metric, value)
print_msg += ' lr={:.6f} step/sec={:.2f} | ETA {}'.format(lr, timer.timing, timer.eta)
......@@ -139,9 +136,7 @@ class CustomTrainer(Trainer):
self.save_model(best_model_path)
self._save_metrics()
metric_msg = [
'{}={:.4f}'.format(metric, value) for metric, value in self.best_metrics.items()
]
metric_msg = ['{}={:.4f}'.format(metric, value) for metric, value in self.best_metrics.items()]
metric_msg = ' '.join(metric_msg)
logger.eval('Saving best model to {} [best {}]'.format(best_model_path, metric_msg))
......
demo/autoaug/search.py
浏览文件 @ d6b6c6ff
from auto_augment.autoaug.experiment.experiment import AutoAugExperiment
from auto_augment.autoaug.utils.yaml_config import get_config
from hub_fitter import HubFitterClassifer
......@@ -9,14 +8,21 @@ logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
parser = argparse.ArgumentParser()
parser.add_argument("--config",help="config file",)
parser.add_argument("--workspace",default=None, help="work_space",)
parser.add_argument(
"--config",
help="config file",
)
parser.add_argument(
"--workspace",
default=None,
help="work_space",
)
def main():
search_test()
def search_test():
args = parser.parse_args()
config = args.config
......@@ -38,14 +44,13 @@ def search_test():
resource_config=resource_config,
data_config=data_config,
search_space=search_space,
fitter=HubFitterClassifer
)
fitter=HubFitterClassifer)
result = exper.search() # 开始搜索任务
policy = result.get_best_policy() # 最佳策略获取， policy格式见 搜索结果应用格式
print("policy is:{}".format(policy))
dump_path = os.path.join(workspace, "auto_aug_config.json")
result.dump_best_policy(
path=dump_path)
result.dump_best_policy(path=dump_path)
if __name__ == "__main__":
main()
demo/autoaug/train.py
浏览文件 @ d6b6c6ff
......@@ -24,9 +24,20 @@ logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
parser = argparse.ArgumentParser()
parser.add_argument("--config",help="config file",)
parser.add_argument("--workspace",default=None, help="work_space",)
parser.add_argument("--policy",default=None, help="data aug policy",)
parser.add_argument(
"--config",
help="config file",
)
parser.add_argument(
"--workspace",
default=None,
help="work_space",
)
parser.add_argument(
"--policy",
default=None,
help="data aug policy",
)
if __name__ == '__main__':
args = parser.parse_args()
......@@ -39,10 +50,7 @@ if __name__ == '__main__':
input_size = task_config.classifier.input_size
scale_size = task_config.classifier.scale_size
normalize = {
'mean': [
0.485, 0.456, 0.406], 'std': [
0.229, 0.224, 0.225]}
normalize = {'mean': [0.485, 0.456, 0.406], 'std': [0.229, 0.224, 0.225]}
epochs = task_config.classifier.epochs
policy = args.policy
......@@ -50,13 +58,11 @@ if __name__ == '__main__':
print("use normal train transform")
TrainTransform = transforms.Compose(
transforms=[
transforms.Resize(
(input_size,
input_size)),
transforms.Resize((input_size, input_size)),
transforms.Permute(),
transforms.Normalize(
**normalize, channel_first=True)],
channel_first = False)
transforms.Normalize(**normalize, channel_first=True)
],
channel_first=False)
else:
TrainTransform = PbaAugment(
input_size=input_size,
......@@ -64,17 +70,12 @@ if __name__ == '__main__':
normalize=normalize,
policy=policy,
hp_policy_epochs=epochs,
stage="train"
)
stage="train")
train_dataset, eval_dataset = _init_loader(config, TrainTransform=TrainTransform)
class_to_id_dict = _read_classes(config.data_config.label_list)
model = hub.Module(name=config.task_config.classifier.model_name, label_list=class_to_id_dict.keys(), load_checkpoint=None)
model = hub.Module(
name=config.task_config.classifier.model_name, label_list=class_to_id_dict.keys(), load_checkpoint=None)
optimizer = paddle.optimizer.Adam(
learning_rate=0.001, parameters=model.parameters())
trainer = CustomTrainer(
model=model,
optimizer=optimizer,
checkpoint_dir='img_classification_ckpt')
optimizer = paddle.optimizer.Adam(learning_rate=0.001, parameters=model.parameters())
trainer = CustomTrainer(model=model, optimizer=optimizer, checkpoint_dir='img_classification_ckpt')
trainer.train(train_dataset, epochs=epochs, batch_size=32, eval_dataset=eval_dataset, save_interval=10)
demo/colorization/train.py
浏览文件 @ d6b6c6ff
......@@ -6,7 +6,8 @@ from paddlehub.datasets import Canvas
if __name__ == '__main__':
transform = T.Compose([T.Resize((256, 256), interpolation='NEAREST'),
transform = T.Compose(
[T.Resize((256, 256), interpolation='NEAREST'),
T.RandomPaddingCrop(crop_size=176),
T.RGB2LAB()], to_rgb=True)
......
demo/semantic_segmentation/README.md
浏览文件 @ d6b6c6ff
......@@ -159,5 +159,3 @@ https://github.com/PaddlePaddle/PaddleSeg
paddlepaddle >= 2.0.0
paddlehub >= 2.0.0
demo/sequence_labeling/predict.py
浏览文件 @ d6b6c6ff
......@@ -24,9 +24,7 @@ if __name__ == '__main__':
'华裔作家韩素音女士曾三次到大足，称“大足石窟是一座未被开发的金矿”。',
]
data = [[split_char.join(text)] for text in text_a]
label_map = {
idx: label for idx, label in enumerate(label_list)
}
label_map = {idx: label for idx, label in enumerate(label_list)}
model = hub.Module(
name='ernie_tiny',
......@@ -39,4 +37,3 @@ if __name__ == '__main__':
results = model.predict(data=data, max_seq_len=128, batch_size=1, use_gpu=True)
for idx, text in enumerate(text_a):
print(f'Text:\n{text} \nLable: \n{", ".join(results[idx][1:len(text)+1])} \n')
demo/sequence_labeling/train.py
浏览文件 @ d6b6c6ff
......@@ -21,7 +21,11 @@ import argparse
parser = argparse.ArgumentParser(__doc__)
parser.add_argument("--num_epoch", type=int, default=3, help="Number of epoches for fine-tuning.")
parser.add_argument("--use_gpu", type=ast.literal_eval, default=True, help="Whether use GPU for fine-tuning, input should be True or False")
parser.add_argument(
"--use_gpu",
type=ast.literal_eval,
default=True,
help="Whether use GPU for fine-tuning, input should be True or False")
parser.add_argument("--learning_rate", type=float, default=5e-5, help="Learning rate used to train with warmup.")
parser.add_argument("--max_seq_len", type=int, default=128, help="Number of words of the longest seqence.")
parser.add_argument("--batch_size", type=int, default=32, help="Total examples' number in batch for training.")
......@@ -30,12 +34,9 @@ parser.add_argument("--save_interval", type=int, default=1, help="Save checkpoin
args = parser.parse_args()
if __name__ == '__main__':
label_list = MSRA_NER.label_list
label_map = {
idx: label for idx, label in enumerate(label_list)
}
label_map = {idx: label for idx, label in enumerate(label_list)}
model = hub.Module(
name='ernie_tiny',
......@@ -45,21 +46,9 @@ if __name__ == '__main__':
)
tokenizer = model.get_tokenizer()
train_dataset = MSRA_NER(
tokenizer=tokenizer,
max_seq_len=args.max_seq_len,
mode='train'
)
dev_dataset = MSRA_NER(
tokenizer=tokenizer,
max_seq_len=args.max_seq_len,
mode='dev'
)
test_dataset = MSRA_NER(
tokenizer=tokenizer,
max_seq_len=args.max_seq_len,
mode='test'
)
train_dataset = MSRA_NER(tokenizer=tokenizer, max_seq_len=args.max_seq_len, mode='train')
dev_dataset = MSRA_NER(tokenizer=tokenizer, max_seq_len=args.max_seq_len, mode='dev')
test_dataset = MSRA_NER(tokenizer=tokenizer, max_seq_len=args.max_seq_len, mode='test')
optimizer = paddle.optimizer.AdamW(learning_rate=args.learning_rate, parameters=model.parameters())
trainer = hub.Trainer(model, optimizer, checkpoint_dir=args.checkpoint_dir, use_gpu=args.use_gpu)
......
demo/style_transfer/predict.py
浏览文件 @ d6b6c6ff
......@@ -3,4 +3,4 @@ import paddlehub as hub
if __name__ == '__main__':
model = hub.Module(name='msgnet', load_checkpoint="/PATH/TO/CHECKPOINT")
result = model.predict(origin=["venice-boat.jpg"], style="candy.jpg", visualization=True, save_path ='style_tranfer')
\ No newline at end of file
result = model.predict(origin=["venice-boat.jpg"], style="candy.jpg", visualization=True, save_path='style_tranfer')
demo/text_classification/README.md
浏览文件 @ d6b6c6ff
......@@ -166,4 +166,3 @@ for idx, text in enumerate(data):
```
参数配置正确后，请执行脚本`python predict.py`， 加载模型具体可参见[加载](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc/api/paddle/framework/io/load_cn.html#load)。
demo/text_classification/embedding/predict.py
浏览文件 @ d6b6c6ff
......@@ -19,17 +19,19 @@ from model import BoWModel
import ast
import argparse
parser = argparse.ArgumentParser(__doc__)
parser.add_argument("--hub_embedding_name", type=str, default='w2v_baidu_encyclopedia_target_word-word_dim300', help="")
parser.add_argument("--max_seq_len", type=int, default=128, help="Number of words of the longest seqence.")
parser.add_argument("--batch_size", type=int, default=64, help="Total examples' number in batch for training.")
parser.add_argument("--checkpoint", type=str, default='./checkpoint/best_model/model.pdparams', help="Model checkpoint")
parser.add_argument("--use_gpu", type=ast.literal_eval, default=True, help="Whether use GPU for fine-tuning, input should be True or False")
parser.add_argument(
"--use_gpu",
type=ast.literal_eval,
default=True,
help="Whether use GPU for fine-tuning, input should be True or False")
args = parser.parse_args()
if __name__ == '__main__':
# Data to be prdicted
data = [
......@@ -44,12 +46,9 @@ if __name__ == '__main__':
embedder = hub.Module(name=args.hub_embedding_name)
tokenizer = embedder.get_tokenizer()
model = BoWModel(
embedder=embedder,
tokenizer=tokenizer,
load_checkpoint=args.checkpoint,
label_map=label_map)
model = BoWModel(embedder=embedder, tokenizer=tokenizer, load_checkpoint=args.checkpoint, label_map=label_map)
results = model.predict(data, max_seq_len=args.max_seq_len, batch_size=args.batch_size, use_gpu=args.use_gpu, return_result=False)
results = model.predict(
data, max_seq_len=args.max_seq_len, batch_size=args.batch_size, use_gpu=args.use_gpu, return_result=False)
for idx, text in enumerate(data):
print('Data: {} \t Lable: {}'.format(text[0], results[idx]))
demo/text_classification/embedding/train.py
浏览文件 @ d6b6c6ff
......@@ -21,7 +21,6 @@ from model import BoWModel
import ast
import argparse
parser = argparse.ArgumentParser(__doc__)
parser.add_argument("--hub_embedding_name", type=str, default='w2v_baidu_encyclopedia_target_word-word_dim300', help="")
parser.add_argument("--num_epoch", type=int, default=10, help="Number of epoches for fine-tuning.")
......@@ -30,11 +29,14 @@ parser.add_argument("--max_seq_len", type=int, default=128, help="Number of word
parser.add_argument("--batch_size", type=int, default=64, help="Total examples' number in batch for training.")
parser.add_argument("--checkpoint_dir", type=str, default='./checkpoint', help="Directory to model checkpoint")
parser.add_argument("--save_interval", type=int, default=5, help="Save checkpoint every n epoch.")
parser.add_argument("--use_gpu", type=ast.literal_eval, default=True, help="Whether use GPU for fine-tuning, input should be True or False")
parser.add_argument(
"--use_gpu",
type=ast.literal_eval,
default=True,
help="Whether use GPU for fine-tuning, input should be True or False")
args = parser.parse_args()
if __name__ == '__main__':
embedder = hub.Module(name=args.hub_embedding_name)
tokenizer = embedder.get_tokenizer()
......@@ -44,8 +46,7 @@ if __name__ == '__main__':
test_dataset = ChnSentiCorp(tokenizer=tokenizer, max_seq_len=args.max_seq_len, mode='test')
model = BoWModel(embedder=embedder)
optimizer = paddle.optimizer.AdamW(
learning_rate=args.learning_rate, parameters=model.parameters())
optimizer = paddle.optimizer.AdamW(learning_rate=args.learning_rate, parameters=model.parameters())
trainer = hub.Trainer(model, optimizer, checkpoint_dir=args.checkpoint_dir, use_gpu=args.use_gpu)
trainer.train(
train_dataset,
......
demo/text_classification/train.py
浏览文件 @ d6b6c6ff
......@@ -21,7 +21,11 @@ import argparse
parser = argparse.ArgumentParser(__doc__)
parser.add_argument("--num_epoch", type=int, default=3, help="Number of epoches for fine-tuning.")
parser.add_argument("--use_gpu", type=ast.literal_eval, default=True, help="Whether use GPU for fine-tuning, input should be True or False")
parser.add_argument(
"--use_gpu",
type=ast.literal_eval,
default=True,
help="Whether use GPU for fine-tuning, input should be True or False")
parser.add_argument("--learning_rate", type=float, default=5e-5, help="Learning rate used to train with warmup.")
parser.add_argument("--max_seq_len", type=int, default=128, help="Number of words of the longest seqence.")
parser.add_argument("--batch_size", type=int, default=32, help="Total examples' number in batch for training.")
......@@ -33,15 +37,11 @@ args = parser.parse_args()
if __name__ == '__main__':
model = hub.Module(name='ernie_tiny', version='2.0.1', task='seq-cls')
train_dataset = ChnSentiCorp(
tokenizer=model.get_tokenizer(), max_seq_len=args.max_seq_len, mode='train')
dev_dataset = ChnSentiCorp(
tokenizer=model.get_tokenizer(), max_seq_len=args.max_seq_len, mode='dev')
test_dataset = ChnSentiCorp(
tokenizer=model.get_tokenizer(), max_seq_len=args.max_seq_len, mode='test')
train_dataset = ChnSentiCorp(tokenizer=model.get_tokenizer(), max_seq_len=args.max_seq_len, mode='train')
dev_dataset = ChnSentiCorp(tokenizer=model.get_tokenizer(), max_seq_len=args.max_seq_len, mode='dev')
test_dataset = ChnSentiCorp(tokenizer=model.get_tokenizer(), max_seq_len=args.max_seq_len, mode='test')
optimizer = paddle.optimizer.AdamW(
learning_rate=args.learning_rate, parameters=model.parameters())
optimizer = paddle.optimizer.AdamW(learning_rate=args.learning_rate, parameters=model.parameters())
trainer = hub.Trainer(model, optimizer, checkpoint_dir=args.checkpoint_dir, use_gpu=args.use_gpu)
trainer.train(
train_dataset,
......
demo/text_matching/train.py
浏览文件 @ d6b6c6ff
......@@ -21,7 +21,11 @@ import argparse
parser = argparse.ArgumentParser(__doc__)
parser.add_argument("--num_epoch", type=int, default=10, help="Number of epoches for fine-tuning.")
parser.add_argument("--use_gpu", type=ast.literal_eval, default=True, help="Whether use GPU for fine-tuning, input should be True or False")
parser.add_argument(
"--use_gpu",
type=ast.literal_eval,
default=True,
help="Whether use GPU for fine-tuning, input should be True or False")
parser.add_argument("--learning_rate", type=float, default=5e-5, help="Learning rate used to train with warmup.")
parser.add_argument("--max_seq_len", type=int, default=64, help="Number of words of the longest seqence.")
parser.add_argument("--batch_size", type=int, default=128, help="Total examples' number in batch for training.")
......@@ -38,8 +42,7 @@ if __name__ == '__main__':
dev_dataset = LCQMC(tokenizer=tokenizer, max_seq_len=args.max_seq_len, mode='dev')
test_dataset = LCQMC(tokenizer=tokenizer, max_seq_len=args.max_seq_len, mode='test')
optimizer = paddle.optimizer.AdamW(
learning_rate=args.learning_rate, parameters=model.parameters())
optimizer = paddle.optimizer.AdamW(learning_rate=args.learning_rate, parameters=model.parameters())
trainer = hub.Trainer(model, optimizer, checkpoint_dir=args.checkpoint_dir, use_gpu=args.use_gpu)
trainer.train(
train_dataset,
......
docs/docs_ch/figures.md
浏览文件 @ d6b6c6ff
......@@ -93,6 +93,3 @@ $ hub serving start -m chinese_ocr_db_crnn_mobile
</p>
* 如需在线快速体验，请点击[PaddleHub教程合集](https://aistudio.baidu.com/aistudio/projectdetail/231146)，可使用AI Studio平台提供的GPU算力进行快速尝试。
modules/audio/audio_classification/PANNs/cnn10/module.py
浏览文件 @ d6b6c6ff
......@@ -27,7 +27,8 @@ from paddlehub.module.module import moduleinfo
from paddlehub.utils.log import logger
@moduleinfo(name="panns_cnn10",
@moduleinfo(
name="panns_cnn10",
version="1.0.0",
summary="",
author="Baidu",
......@@ -52,15 +53,15 @@ class PANN(nn.Layer):
self.num_class = num_class
if task == 'sound-cls':
self.cnn10 = CNN10(extract_embedding=True,
checkpoint=os.path.join(MODULE_HOME, 'panns_cnn10', 'cnn10.pdparams'))
self.cnn10 = CNN10(
extract_embedding=True, checkpoint=os.path.join(MODULE_HOME, 'panns_cnn10', 'cnn10.pdparams'))
self.dropout = nn.Dropout(0.1)
self.fc = nn.Linear(self.cnn10.emb_size, num_class)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
else:
self.cnn10 = CNN10(extract_embedding=False,
checkpoint=os.path.join(MODULE_HOME, 'panns_cnn10', 'cnn10.pdparams'))
self.cnn10 = CNN10(
extract_embedding=False, checkpoint=os.path.join(MODULE_HOME, 'panns_cnn10', 'cnn10.pdparams'))
self.task = task
if load_checkpoint is not None and os.path.isfile(load_checkpoint):
......
modules/audio/audio_classification/PANNs/cnn10/network.py
浏览文件 @ d6b6c6ff
......@@ -25,13 +25,15 @@ class ConvBlock(nn.Layer):
def __init__(self, in_channels, out_channels):
super(ConvBlock, self).__init__()
self.conv1 = nn.Conv2D(in_channels=in_channels,
self.conv1 = nn.Conv2D(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=(3, 3),
stride=(1, 1),
padding=(1, 1),
bias_attr=False)
self.conv2 = nn.Conv2D(in_channels=out_channels,
self.conv2 = nn.Conv2D(
in_channels=out_channels,
out_channels=out_channels,
kernel_size=(3, 3),
stride=(1, 1),
......
modules/audio/audio_classification/PANNs/cnn14/module.py
浏览文件 @ d6b6c6ff
......@@ -27,7 +27,8 @@ from paddlehub.module.module import moduleinfo
from paddlehub.utils.log import logger
@moduleinfo(name="panns_cnn14",
@moduleinfo(
name="panns_cnn14",
version="1.0.0",
summary="",
author="Baidu",
......@@ -52,15 +53,15 @@ class PANN(nn.Layer):
self.num_class = num_class
if task == 'sound-cls':
self.cnn14 = CNN14(extract_embedding=True,
checkpoint=os.path.join(MODULE_HOME, 'panns_cnn14', 'cnn14.pdparams'))
self.cnn14 = CNN14(
extract_embedding=True, checkpoint=os.path.join(MODULE_HOME, 'panns_cnn14', 'cnn14.pdparams'))
self.dropout = nn.Dropout(0.1)
self.fc = nn.Linear(self.cnn14.emb_size, num_class)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
else:
self.cnn14 = CNN14(extract_embedding=False,
checkpoint=os.path.join(MODULE_HOME, 'panns_cnn14', 'cnn14.pdparams'))
self.cnn14 = CNN14(
extract_embedding=False, checkpoint=os.path.join(MODULE_HOME, 'panns_cnn14', 'cnn14.pdparams'))
self.task = task
if load_checkpoint is not None and os.path.isfile(load_checkpoint):
......
modules/audio/audio_classification/PANNs/cnn14/network.py
浏览文件 @ d6b6c6ff
......@@ -25,13 +25,15 @@ class ConvBlock(nn.Layer):
def __init__(self, in_channels, out_channels):
super(ConvBlock, self).__init__()
self.conv1 = nn.Conv2D(in_channels=in_channels,
self.conv1 = nn.Conv2D(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=(3, 3),
stride=(1, 1),
padding=(1, 1),
bias_attr=False)
self.conv2 = nn.Conv2D(in_channels=out_channels,
self.conv2 = nn.Conv2D(
in_channels=out_channels,
out_channels=out_channels,
kernel_size=(3, 3),
stride=(1, 1),
......
modules/audio/audio_classification/PANNs/cnn6/module.py
浏览文件 @ d6b6c6ff
......@@ -27,7 +27,8 @@ from paddlehub.module.module import moduleinfo
from paddlehub.utils.log import logger
@moduleinfo(name="panns_cnn6",
@moduleinfo(
name="panns_cnn6",
version="1.0.0",
summary="",
author="Baidu",
......@@ -52,15 +53,15 @@ class PANN(nn.Layer):
self.num_class = num_class
if task == 'sound-cls':
self.cnn6 = CNN6(extract_embedding=True,
checkpoint=os.path.join(MODULE_HOME, 'panns_cnn6', 'cnn6.pdparams'))
self.cnn6 = CNN6(
extract_embedding=True, checkpoint=os.path.join(MODULE_HOME, 'panns_cnn6', 'cnn6.pdparams'))
self.dropout = nn.Dropout(0.1)
self.fc = nn.Linear(self.cnn6.emb_size, num_class)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
else:
self.cnn6 = CNN6(extract_embedding=False,
checkpoint=os.path.join(MODULE_HOME, 'panns_cnn6', 'cnn6.pdparams'))
self.cnn6 = CNN6(
extract_embedding=False, checkpoint=os.path.join(MODULE_HOME, 'panns_cnn6', 'cnn6.pdparams'))
self.task = task
if load_checkpoint is not None and os.path.isfile(load_checkpoint):
......
modules/audio/audio_classification/PANNs/cnn6/network.py
浏览文件 @ d6b6c6ff
......@@ -25,7 +25,8 @@ class ConvBlock5x5(nn.Layer):
def __init__(self, in_channels, out_channels):
super(ConvBlock5x5, self).__init__()
self.conv1 = nn.Conv2D(in_channels=in_channels,
self.conv1 = nn.Conv2D(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=(5, 5),
stride=(1, 1),
......
modules/image/Image_editing/colorization/user_guided_colorization/data_feed.py
浏览文件 @ d6b6c6ff
......@@ -18,6 +18,7 @@ class ColorizeHint:
hint(np.ndarray): hint images
mask(np.ndarray): mask images
"""
def __init__(self, percent: float, num_points: int = None, samp: str = 'normal', use_avg: bool = True):
self.percent = percent
self.num_points = num_points
......@@ -52,11 +53,9 @@ class ColorizeHint:
# add color point
if self.use_avg:
# embed()
hint[nn, :, h:h + P, w:w + P] = np.mean(np.mean(data[nn, :, h:h + P, w:w + P],
axis=2,
keepdims=True),
axis=1,
keepdims=True).reshape(1, C, 1, 1)
hint[nn, :, h:h + P, w:w + P] = np.mean(
np.mean(data[nn, :, h:h + P, w:w + P], axis=2, keepdims=True), axis=1, keepdims=True).reshape(
1, C, 1, 1)
else:
hint[nn, :, h:h + P, w:w + P] = data[nn, :, h:h + P, w:w + P]
mask[nn, :, h:h + P, w:w + P] = 1
......@@ -82,6 +81,7 @@ class ColorizePreprocess:
data(dict)：The preprocessed data for colorization.
"""
def __init__(self,
ab_thresh: float = 0.,
p: float = 0.,
......@@ -113,8 +113,8 @@ class ColorizePreprocess:
data['B'] = data_lab[:, 1:, :, :]
if self.ab_thresh > 0: # mask out grayscale images
thresh = 1. * self.ab_thresh / 110
mask = np.sum(np.abs(np.max(np.max(data['B'], axis=3), axis=2) - np.min(np.min(data['B'], axis=3), axis=2)),
axis=1)
mask = np.sum(
np.abs(np.max(np.max(data['B'], axis=3), axis=2) - np.min(np.min(data['B'], axis=3), axis=2)), axis=1)
mask = (mask >= thresh)
data['A'] = data['A'][mask, :, :, :]
data['B'] = data['B'][mask, :, :, :]
......
modules/image/Image_editing/colorization/user_guided_colorization/module.py
浏览文件 @ d6b6c6ff
......@@ -40,6 +40,7 @@ class UserGuidedColorization(nn.Layer):
load_checkpoint (str): Pretrained checkpoint path.
"""
def __init__(self, use_tanh: bool = True, load_checkpoint: str = None):
super(UserGuidedColorization, self).__init__()
self.input_nc = 4
......
modules/image/Image_gan/style_transfer/msgnet/module.py
浏览文件 @ d6b6c6ff
......@@ -14,6 +14,7 @@ from paddlehub.module.cv_module import StyleTransferModule
class GramMatrix(nn.Layer):
"""Calculate gram matrix"""
def forward(self, y):
(b, ch, h, w) = y.shape
features = y.reshape((b, ch, w * h))
......@@ -24,6 +25,7 @@ class GramMatrix(nn.Layer):
class ConvLayer(nn.Layer):
"""Basic conv layer with reflection padding layer"""
def __init__(self, in_channels: int, out_channels: int, kernel_size: int, stride: int):
super(ConvLayer, self).__init__()
pad = int(np.floor(kernel_size / 2))
......@@ -51,6 +53,7 @@ class UpsampleConvLayer(nn.Layer):
Return:
img(paddle.Tensor): UpsampleConvLayer output.
"""
def __init__(self, in_channels: int, out_channels: int, kernel_size: int, stride: int, upsample=None):
super(UpsampleConvLayer, self).__init__()
self.upsample = upsample
......@@ -85,6 +88,7 @@ class Bottleneck(nn.Layer):
Return:
img(paddle.Tensor): Bottleneck output.
"""
def __init__(self,
inplanes: int,
planes: int,
......@@ -98,8 +102,8 @@ class Bottleneck(nn.Layer):
self.residual_layer = nn.Conv2D(inplanes, planes * self.expansion, kernel_size=1, stride=stride)
conv_block = (norm_layer(inplanes), nn.ReLU(), nn.Conv2D(inplanes, planes, kernel_size=1, stride=1),
norm_layer(planes), nn.ReLU(), ConvLayer(planes, planes, kernel_size=3, stride=stride),
norm_layer(planes), nn.ReLU(), nn.Conv2D(planes, planes * self.expansion, kernel_size=1,
stride=1))
norm_layer(planes), nn.ReLU(), nn.Conv2D(
planes, planes * self.expansion, kernel_size=1, stride=1))
self.conv_block = nn.Sequential(*conv_block)
def forward(self, x: paddle.Tensor):
......@@ -125,14 +129,12 @@ class UpBottleneck(nn.Layer):
Return:
img(paddle.Tensor): UpBottleneck output.
"""
def __init__(self, inplanes: int, planes: int, stride: int = 2, norm_layer: nn.Layer = nn.BatchNorm2D):
super(UpBottleneck, self).__init__()
self.expansion = 4
self.residual_layer = UpsampleConvLayer(inplanes,
planes * self.expansion,
kernel_size=1,
stride=1,
upsample=stride)
self.residual_layer = UpsampleConvLayer(
inplanes, planes * self.expansion, kernel_size=1, stride=1, upsample=stride)
conv_block = []
conv_block += [norm_layer(inplanes), nn.ReLU(), nn.Conv2D(inplanes, planes, kernel_size=1, stride=1)]
conv_block += [
......@@ -163,6 +165,7 @@ class Inspiration(nn.Layer):
Return:
img(paddle.Tensor): UpBottleneck output.
"""
def __init__(self, C: int, B: int = 1):
super(Inspiration, self).__init__()
......@@ -179,8 +182,8 @@ class Inspiration(nn.Layer):
self.P = paddle.bmm(self.weight.expand_as(self.G), self.G)
x = paddle.bmm(
self.P.transpose((0, 2, 1)).expand((X.shape[0], self.C, self.C)), X.reshape(
(X.shape[0], X.shape[1], -1))).reshape(X.shape)
self.P.transpose((0, 2, 1)).expand((X.shape[0], self.C, self.C)), X.reshape((X.shape[0], X.shape[1],
-1))).reshape(X.shape)
return x
def __repr__(self):
......@@ -190,6 +193,7 @@ class Inspiration(nn.Layer):
class Vgg16(nn.Layer):
""" First four layers from Vgg16."""
def __init__(self):
super(Vgg16, self).__init__()
self.conv1_1 = nn.Conv2D(3, 64, kernel_size=3, stride=1, padding=1)
......@@ -264,12 +268,8 @@ class MSGNet(nn.Layer):
Return:
img(paddle.Tensor): MSGNet output.
"""
def __init__(self,
input_nc=3,
output_nc=3,
ngf=128,
n_blocks=6,
norm_layer=nn.InstanceNorm2D,
def __init__(self, input_nc=3, output_nc=3, ngf=128, n_blocks=6, norm_layer=nn.InstanceNorm2D,
load_checkpoint=None):
super(MSGNet, self).__init__()
self.gram = GramMatrix()
......
modules/image/classification/resnet50_vd_imagenet_ssld/module.py
浏览文件 @ d6b6c6ff
......@@ -194,7 +194,7 @@ class ResNet50_vd(nn.Layer):
label_file = os.path.join(self.directory, 'label_list.txt')
files = open(label_file)
for line in files.readlines():
line=line.strip('\n')
line = line.strip('\n')
label_list.append(line)
self.labels = label_list
class_dim = len(self.labels)
......@@ -248,7 +248,12 @@ class ResNet50_vd(nn.Layer):
print("load pretrained checkpoint success")
def transforms(self, images: Union[str, np.ndarray]):
transforms = T.Compose([T.Resize((256, 256)), T.CenterCrop(224), T.Normalize(mean=[0.485, 0.456, 0.406], std = [0.229, 0.224, 0.225])], to_rgb=True)
transforms = T.Compose([
T.Resize((256, 256)),
T.CenterCrop(224),
T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
],
to_rgb=True)
return transforms(images)
def forward(self, inputs: paddle.Tensor):
......
modules/image/industrial_application/meter_readings/barometer_reader/README.md
浏览文件 @ d6b6c6ff
......@@ -111,5 +111,3 @@ paddlepaddle >= 2.0.0
paddlehub >= 2.0.0
paddlex >= 1.3.0
modules/image/industrial_application/meter_readings/barometer_reader/module.py
浏览文件 @ d6b6c6ff
......@@ -12,7 +12,6 @@ import paddle.nn as nn
from paddlex.seg import transforms as T
from paddlehub.module.module import moduleinfo, runnable, serving
METER_SHAPE = 512
CIRCLE_CENTER = [256, 256]
CIRCLE_RADIUS = 250
......@@ -44,11 +43,13 @@ def cv2_to_base64(image: np.ndarray):
return base64.b64encode(data.tostring()).decode('utf8')
@moduleinfo(name="barometer_reader",
@moduleinfo(
name="barometer_reader",
type="CV/image_editing",
author="paddlepaddle",
author_email="",
summary="meter_reader implements the detection and automatic reading of traditional mechanical pointer meters based on Meter detection and pointer segmentation.",
summary=
"meter_reader implements the detection and automatic reading of traditional mechanical pointer meters based on Meter detection and pointer segmentation.",
version="1.0.0")
class BarometerReader(nn.Layer):
def __init__(self):
......@@ -121,8 +122,7 @@ class BarometerReader(nn.Layer):
if pointer_flag:
if pointer_data[i] == 0 and pointer_data[i + 1] == 0:
one_pointer_end = i - 1
pointer_location = (
one_pointer_start + one_pointer_end) / 2
pointer_location = (one_pointer_start + one_pointer_end) / 2
one_pointer_start = 0
one_pointer_end = 0
pointer_flag = False
......@@ -135,8 +135,7 @@ class BarometerReader(nn.Layer):
if scale_location[i] <= pointer_location and pointer_location < scale_location[i + 1]:
scales = i + (pointer_location - scale_location[i]) / (
scale_location[i + 1] - scale_location[i] + 1e-05) + 1
ratio = (pointer_location - scale_location[0]) / (
scale_location[scale_num - 1] - scale_location[0] + 1e-05)
ratio = (pointer_location - scale_location[0]) / (scale_location[scale_num - 1] - scale_location[0] + 1e-05)
result = {'scale_num': scale_num, 'scales': scales, 'ratio': ratio}
return result
......@@ -147,7 +146,7 @@ class BarometerReader(nn.Layer):
erode_kernel: int = 4,
use_erode: bool = True,
visualization: bool = False,
save_dir: str ='output'):
save_dir: str = 'output'):
if isinstance(im_file, str):
im = cv2.imread(im_file).astype('float32')
......@@ -173,13 +172,7 @@ class BarometerReader(nn.Layer):
meter_shape = sub_image.shape
scale_x = float(METER_SHAPE) / float(meter_shape[1])
scale_y = float(METER_SHAPE) / float(meter_shape[0])
meter_meter = cv2.resize(
sub_image,
None,
None,
fx=scale_x,
fy=scale_y,
interpolation=cv2.INTER_LINEAR)
meter_meter = cv2.resize(sub_image, None, None, fx=scale_x, fy=scale_y, interpolation=cv2.INTER_LINEAR)
meter_meter = meter_meter.astype('float32')
resized_meters.append(meter_meter)
......@@ -191,9 +184,7 @@ class BarometerReader(nn.Layer):
for j in range(i, im_size):
meter_images.append(resized_meters[j - i])
result = self.segmenter.batch_predict(
transforms=self.seg_transform,
img_file_list=meter_images)
result = self.segmenter.batch_predict(transforms=self.seg_transform, img_file_list=meter_images)
if use_erode:
kernel = np.ones((erode_kernel, erode_kernel), np.uint8)
......@@ -250,21 +241,16 @@ class BarometerReader(nn.Layer):
prog='hub run {}'.format(self.name),
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, not required.")
title="Config options", description="Run configuration for controlling module behavior, not required.")
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
results = self.predict(im_file=args.input_path)
return results
def add_module_input_arg(self):
"""
Add the command input options.
"""
self.arg_input_group.add_argument(
'--input_path', type=str, help="path to image.")
self.arg_input_group.add_argument('--input_path', type=str, help="path to image.")
modules/image/instance_segmentation/solov2/README.md
浏览文件 @ d6b6c6ff
......@@ -116,4 +116,3 @@ https://github.com/PaddlePaddle/PaddleDetection
paddlepaddle >= 2.0.0
paddlehub >= 2.0.0
modules/image/instance_segmentation/solov2/data_feed.py
浏览文件 @ d6b6c6ff
......@@ -42,38 +42,28 @@ def visualize_box_mask(im, results, labels=None, mask_resolution=14, threshold=0
im (PIL.Image.Image): visualized image
"""
if not labels:
labels = ['background', 'person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus',
'train', 'truck', 'boat', 'traffic light', 'fire', 'hydrant', 'stop sign', 'parking meter',
'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe',
'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball',
'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard', 'tennis racket', 'bottle',
'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple', 'sandwich', 'orange',
'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch', 'potted plant', 'bed',
'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 'cell phone', 'microwave',
'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 'teddy bear',
'hair drier', 'toothbrush']
labels = [
'background', 'person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat',
'traffic light', 'fire', 'hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse',
'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie',
'suitcase', 'frisbee', 'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove',
'skateboard', 'surfboard', 'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl',
'banana', 'apple', 'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair',
'couch', 'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard',
'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors',
'teddy bear', 'hair drier', 'toothbrush'
]
if isinstance(im, str):
im = Image.open(im).convert('RGB')
else:
im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
im = Image.fromarray(im)
if 'masks' in results and 'boxes' in results:
im = draw_mask(
im,
results['boxes'],
results['masks'],
labels,
resolution=mask_resolution)
im = draw_mask(im, results['boxes'], results['masks'], labels, resolution=mask_resolution)
if 'boxes' in results:
im = draw_box(im, results['boxes'], labels)
if 'segm' in results:
im = draw_segm(
im,
results['segm'],
results['label'],
results['score'],
labels,
threshold=threshold)
im = draw_segm(im, results['segm'], results['label'], results['score'], labels, threshold=threshold)
return im
......@@ -200,28 +190,19 @@ def draw_box(im, np_boxes, labels):
color = tuple(clsid2color[clsid])
# draw bbox
draw.line(
[(xmin, ymin), (xmin, ymax), (xmax, ymax), (xmax, ymin),
(xmin, ymin)],
draw.line([(xmin, ymin), (xmin, ymax), (xmax, ymax), (xmax, ymin), (xmin, ymin)],
width=draw_thickness,
fill=color)
# draw label
text = "{} {:.4f}".format(labels[clsid], score)
tw, th = draw.textsize(text)
draw.rectangle(
[(xmin + 1, ymin - th), (xmin + tw + 1, ymin)], fill=color)
draw.rectangle([(xmin + 1, ymin - th), (xmin + tw + 1, ymin)], fill=color)
draw.text((xmin + 1, ymin - th), text, fill=(255, 255, 255))
return im
def draw_segm(im,
np_segms,
np_label,
np_score,
labels,
threshold=0.5,
alpha=0.7):
def draw_segm(im, np_segms, np_label, np_score, labels, threshold=0.5, alpha=0.7):
"""
Draw segmentation on image.
"""
......@@ -250,28 +231,17 @@ def draw_segm(im,
sum_y = np.sum(mask, axis=1)
y = np.where(sum_y > 0.5)[0]
x0, x1, y0, y1 = x[0], x[-1], y[0], y[-1]
cv2.rectangle(im, (x0, y0), (x1, y1),
tuple(color_mask.astype('int32').tolist()), 1)
cv2.rectangle(im, (x0, y0), (x1, y1), tuple(color_mask.astype('int32').tolist()), 1)
bbox_text = '%s %.2f' % (labels[clsid], score)
t_size = cv2.getTextSize(bbox_text, 0, 0.3, thickness=1)[0]
cv2.rectangle(im, (x0, y0), (x0 + t_size[0], y0 - t_size[1] - 3),
tuple(color_mask.astype('int32').tolist()), -1)
cv2.putText(
im,
bbox_text, (x0, y0 - 2),
cv2.FONT_HERSHEY_SIMPLEX,
0.3, (0, 0, 0),
1,
lineType=cv2.LINE_AA)
cv2.rectangle(im, (x0, y0), (x0 + t_size[0], y0 - t_size[1] - 3), tuple(color_mask.astype('int32').tolist()),
-1)
cv2.putText(im, bbox_text, (x0, y0 - 2), cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 0), 1, lineType=cv2.LINE_AA)
return Image.fromarray(im.astype('uint8'))
def load_predictor(model_dir,
run_mode='fluid',
batch_size=1,
use_gpu=False,
min_subgraph_size=3):
def load_predictor(model_dir, run_mode='fluid', batch_size=1, use_gpu=False, min_subgraph_size=3):
"""set AnalysisConfig, generate AnalysisPredictor
Args:
model_dir (str): root path of __model__ and __params__
......@@ -282,20 +252,16 @@ def load_predictor(model_dir,
ValueError: predict by TensorRT need use_gpu == True.
"""
if not use_gpu and not run_mode == 'fluid':
raise ValueError(
"Predict by TensorRT mode: {}, expect use_gpu==True, but use_gpu == {}"
.format(run_mode, use_gpu))
raise ValueError("Predict by TensorRT mode: {}, expect use_gpu==True, but use_gpu == {}".format(
run_mode, use_gpu))
if run_mode == 'trt_int8':
raise ValueError("TensorRT int8 mode is not supported now, "
"please use trt_fp32 or trt_fp16 instead.")
raise ValueError("TensorRT int8 mode is not supported now, " "please use trt_fp32 or trt_fp16 instead.")
precision_map = {
'trt_int8': fluid.core.AnalysisConfig.Precision.Int8,
'trt_fp32': fluid.core.AnalysisConfig.Precision.Float32,
'trt_fp16': fluid.core.AnalysisConfig.Precision.Half
}
config = fluid.core.AnalysisConfig(
os.path.join(model_dir, '__model__'),
os.path.join(model_dir, '__params__'))
config = fluid.core.AnalysisConfig(os.path.join(model_dir, '__model__'), os.path.join(model_dir, '__params__'))
if use_gpu:
# initial GPU memory(M), device ID
config.enable_use_gpu(100, 0)
......
modules/image/instance_segmentation/solov2/module.py
浏览文件 @ d6b6c6ff
......@@ -33,20 +33,16 @@ class Detector(object):
threshold (float): threshold to reserve the result for output.
"""
def __init__(self,
min_subgraph_size: int = 60,
use_gpu=False,
threshold: float = 0.5):
def __init__(self, min_subgraph_size: int = 60, use_gpu=False, threshold: float = 0.5):
model_dir = os.path.join(self.directory, 'solov2_r50_fpn_1x')
self.predictor = D.load_predictor(
model_dir,
min_subgraph_size=min_subgraph_size,
use_gpu=use_gpu)
self.compose = [P.Resize(max_size=1333),
self.predictor = D.load_predictor(model_dir, min_subgraph_size=min_subgraph_size, use_gpu=use_gpu)
self.compose = [
P.Resize(max_size=1333),
P.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
P.Permute(),
P.PadStride(stride=32)]
P.PadStride(stride=32)
]
def transform(self, im: Union[str, np.ndarray]):
im, im_info = P.preprocess(im, self.compose)
......@@ -61,17 +57,14 @@ class Detector(object):
for box in np_boxes:
print('class_id:{:d}, confidence:{:.4f},'
'left_top:[{:.2f},{:.2f}],'
' right_bottom:[{:.2f},{:.2f}]'.format(
int(box[0]), box[1], box[2], box[3], box[4], box[5]))
' right_bottom:[{:.2f},{:.2f}]'.format(int(box[0]), box[1], box[2], box[3], box[4], box[5]))
results['boxes'] = np_boxes
if np_masks is not None:
np_masks = np_masks[expect_boxes, :, :, :]
results['masks'] = np_masks
return results
def predict(self,
image: Union[str, np.ndarray],
threshold: float = 0.5):
def predict(self, image: Union[str, np.ndarray], threshold: float = 0.5):
'''
Args:
image (str/np.ndarray): path of image/ np.ndarray read by cv2
......@@ -117,12 +110,9 @@ class DetectorSOLOv2(Detector):
use_gpu (bool): whether use gpu
threshold (float): threshold to reserve the result for output.
"""
def __init__(self,
use_gpu: bool = False,
threshold: float = 0.5):
super(DetectorSOLOv2, self).__init__(
use_gpu=use_gpu,
threshold=threshold)
def __init__(self, use_gpu: bool = False, threshold: float = 0.5):
super(DetectorSOLOv2, self).__init__(use_gpu=use_gpu, threshold=threshold)
def predict(self,
image: Union[str, np.ndarray],
......@@ -148,12 +138,9 @@ class DetectorSOLOv2(Detector):
self.predictor.zero_copy_run()
output_names = self.predictor.get_output_names()
np_label = self.predictor.get_output_tensor(output_names[
0]).copy_to_cpu()
np_score = self.predictor.get_output_tensor(output_names[
1]).copy_to_cpu()
np_segms = self.predictor.get_output_tensor(output_names[
2]).copy_to_cpu()
np_label = self.predictor.get_output_tensor(output_names[0]).copy_to_cpu()
np_score = self.predictor.get_output_tensor(output_names[1]).copy_to_cpu()
np_segms = self.predictor.get_output_tensor(output_names[2]).copy_to_cpu()
output = dict(segm=np_segms, label=np_label, score=np_score)
if visualization:
......
modules/image/instance_segmentation/solov2/processor.py
浏览文件 @ d6b6c6ff
......@@ -78,19 +78,12 @@ class Resize(object):
im_channel = im.shape[2]
im_scale_x, im_scale_y = self.generate_scale(im)
if self.use_cv2:
im = cv2.resize(
im,
None,
None,
fx=im_scale_x,
fy=im_scale_y,
interpolation=self.interp)
im = cv2.resize(im, None, None, fx=im_scale_x, fy=im_scale_y, interpolation=self.interp)
else:
resize_w = int(im_scale_x * float(im.shape[1]))
resize_h = int(im_scale_y * float(im.shape[0]))
if self.max_size != 0:
raise TypeError(
'If you set max_size to cap the maximum size of image,'
raise TypeError('If you set max_size to cap the maximum size of image,'
'please set use_cv2 to True to resize the image.')
im = im.astype('uint8')
im = Image.fromarray(im)
......@@ -99,8 +92,7 @@ class Resize(object):
# padding im when image_shape fixed by infer_cfg.yml
if self.max_size != 0 and self.image_shape is not None:
padding_im = np.zeros(
(self.max_size, self.max_size, im_channel), dtype=np.float32)
padding_im = np.zeros((self.max_size, self.max_size, im_channel), dtype=np.float32)
im_h, im_w = im.shape[:2]
padding_im[:im_h, :im_w, :] = im
im = padding_im
......
modules/image/keypoint_detection/openpose_body_estimation/module.py
浏览文件 @ d6b6c6ff
......@@ -207,10 +207,7 @@ class BodyPoseModel(nn.Layer):
save_path = os.path.join(save_path, img_name)
cv2.imwrite(save_path, canvas)
results = {
'candidate': candidate,
'subset': subset,
'data': canvas}
results = {'candidate': candidate, 'subset': subset, 'data': canvas}
return results
......@@ -221,7 +218,7 @@ class BodyPoseModel(nn.Layer):
"""
images_decode = [P.base64_to_cv2(image) for image in images]
results = self.predict(img=images_decode[0], **kwargs)
final={}
final = {}
final['candidate'] = P.cv2_to_base64(results['candidate'])
final['subset'] = P.cv2_to_base64(results['subset'])
final['data'] = P.cv2_to_base64(results['data'])
......@@ -238,19 +235,13 @@ class BodyPoseModel(nn.Layer):
prog='hub run {}'.format(self.name),
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, not required.")
title="Config options", description="Run configuration for controlling module behavior, not required.")
self.add_module_config_arg()
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
results = self.predict(
img=args.input_path,
save_path=args.output_dir,
visualization=args.visualization)
results = self.predict(img=args.input_path, save_path=args.output_dir, visualization=args.visualization)
return results
......@@ -260,19 +251,12 @@ class BodyPoseModel(nn.Layer):
"""
self.arg_config_group.add_argument(
'--output_dir',
type=str,
default='openpose_body',
help="The directory to save output images.")
'--output_dir', type=str, default='openpose_body', help="The directory to save output images.")
self.arg_config_group.add_argument(
'--visualization',
type=bool,
default=True,
help="whether to save output as images.")
'--visualization', type=bool, default=True, help="whether to save output as images.")
def add_module_input_arg(self):
"""
Add the command input options.
"""
self.arg_input_group.add_argument(
'--input_path', type=str, help="path to image.")
\ No newline at end of file
self.arg_input_group.add_argument('--input_path', type=str, help="path to image.")
modules/image/keypoint_detection/openpose_hands_estimation/module.py
浏览文件 @ d6b6c6ff
......@@ -170,7 +170,11 @@ class HandPoseModel(nn.Layer):
return np.array(all_peaks)
def predict(self, img: Union[str, np.ndarray], save_path: str = 'openpose_hand', scale: list = [0.5, 1.0, 1.5, 2.0], visualization: bool = True):
def predict(self,
img: Union[str, np.ndarray],
save_path: str = 'openpose_hand',
scale: list = [0.5, 1.0, 1.5, 2.0],
visualization: bool = True):
self.eval()
self.visualization = visualization
if isinstance(img, str):
......@@ -193,7 +197,11 @@ class HandPoseModel(nn.Layer):
peaks[:, 1] = np.where(peaks[:, 1] == 0, peaks[:, 1], peaks[:, 1] + y)
all_hand_peaks.append(peaks)
canvas = copy.deepcopy(org_img)
canvas = self.draw_pose(canvas, body_result['candidate'], body_result['subset'],)
canvas = self.draw_pose(
canvas,
body_result['candidate'],
body_result['subset'],
)
canvas = self.draw_hand(canvas, all_hand_peaks)
if self.visualization:
if not os.path.exists(save_path):
......@@ -202,9 +210,7 @@ class HandPoseModel(nn.Layer):
save_path = os.path.join(save_path, img_name)
cv2.imwrite(save_path, canvas)
results = {
'all_hand_peaks': all_hand_peaks,
'data': canvas}
results = {'all_hand_peaks': all_hand_peaks, 'data': canvas}
return results
......@@ -215,8 +221,8 @@ class HandPoseModel(nn.Layer):
"""
images_decode = [P.base64_to_cv2(image) for image in images]
results = self.predict(img=images_decode[0], **kwargs)
final={}
final['all_hand_peaks']=[peak.tolist() for peak in results['all_hand_peaks']]
final = {}
final['all_hand_peaks'] = [peak.tolist() for peak in results['all_hand_peaks']]
final['data'] = P.cv2_to_base64(results['data'])
return final
......@@ -230,20 +236,14 @@ class HandPoseModel(nn.Layer):
prog='hub run {}'.format(self.name),
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, not required.")
title="Config options", description="Run configuration for controlling module behavior, not required.")
self.add_module_config_arg()
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
results = self.predict(
img=args.input_path,
save_path=args.output_dir,
scale=args.scale,
visualization=args.visualization)
img=args.input_path, save_path=args.output_dir, scale=args.scale, visualization=args.visualization)
return results
......@@ -253,24 +253,14 @@ class HandPoseModel(nn.Layer):
"""
self.arg_config_group.add_argument(
'--output_dir',
type=str,
default='openpose_hand',
help="The directory to save output images.")
'--output_dir', type=str, default='openpose_hand', help="The directory to save output images.")
self.arg_config_group.add_argument(
'--scale',
type=list,
default=[0.5, 1.0, 1.5, 2.0],
help="The search scale for openpose hands model.")
'--scale', type=list, default=[0.5, 1.0, 1.5, 2.0], help="The search scale for openpose hands model.")
self.arg_config_group.add_argument(
'--visualization',
type=bool,
default=True,
help="whether to save output as images.")
'--visualization', type=bool, default=True, help="whether to save output as images.")
def add_module_input_arg(self):
"""
Add the command input options.
"""
self.arg_input_group.add_argument(
'--input_path', type=str, help="path to image.")
\ No newline at end of file
self.arg_input_group.add_argument('--input_path', type=str, help="path to image.")
modules/image/keypoint_detection/openpose_hands_estimation/processor.py
浏览文件 @ d6b6c6ff
......@@ -232,6 +232,7 @@ class ResizeScaling:
resize_img = cv2.resize(img, (0, 0), fx=scale, fy=scale, interpolation=self.interpolation)
return resize_img
def npmax(array: np.ndarray):
"""Get max value and index."""
arrayindex = array.argmax(1)
......
modules/image/semantic_segmentation/ExtremeC3_Portrait_Segmentation/module.py
浏览文件 @ d6b6c6ff
......@@ -6,6 +6,7 @@ import numpy as np
from paddle.nn import Layer
from paddlehub.module.module import moduleinfo
@moduleinfo(
name="ExtremeC3_Portrait_Segmentation", # 模型名称
type="CV/semantic_segmentation", # 模型类型
......@@ -26,7 +27,7 @@ class ExtremeC3_Portrait_Segmentation(Layer):
self.model.eval()
# 均值方差
self.mean = [107.304565, 115.69884, 132.35703 ]
self.mean = [107.304565, 115.69884, 132.35703]
self.std = [63.97182, 65.1337, 68.29726]
# 读取数据函数
......@@ -107,20 +108,12 @@ class ExtremeC3_Portrait_Segmentation(Layer):
cv2.imwrite(os.path.join(output_dir, 'result_mask_%d.png' % i), mask)
cv2.imwrite(os.path.join(output_dir, 'result_%d.png' % i), result)
results.append({
'mask': mask,
'result': result
})
results.append({'mask': mask, 'result': result})
return results
# 关键点检测函数
def Segmentation(self,
images=None,
paths=None,
batch_size=1,
output_dir='output',
visualization=False):
def Segmentation(self, images=None, paths=None, batch_size=1, output_dir='output', visualization=False):
# 加载数据处理器
datas = self.load_datas(paths, images)
......
modules/image/semantic_segmentation/SINet_Portrait_Segmentation/module.py
浏览文件 @ d6b6c6ff
......@@ -6,6 +6,7 @@ import numpy as np
from paddle.nn import Layer
from paddlehub.module.module import moduleinfo
@moduleinfo(
name="SINet_Portrait_Segmentation", # 模型名称
type="CV/semantic_segmentation", # 模型类型
......@@ -26,7 +27,7 @@ class SINet_Portrait_Segmentation(Layer):
self.model.eval()
# 均值方差
self.mean = [107.304565, 115.69884, 132.35703 ]
self.mean = [107.304565, 115.69884, 132.35703]
self.std = [63.97182, 65.1337, 68.29726]
# 读取数据函数
......@@ -107,20 +108,12 @@ class SINet_Portrait_Segmentation(Layer):
cv2.imwrite(os.path.join(output_dir, 'result_mask_%d.png' % i), mask)
cv2.imwrite(os.path.join(output_dir, 'result_%d.png' % i), result)
results.append({
'mask': mask,
'result': result
})
results.append({'mask': mask, 'result': result})
return results
# 关键点检测函数
def Segmentation(self,
images=None,
paths=None,
batch_size=1,
output_dir='output',
visualization=False):
def Segmentation(self, images=None, paths=None, batch_size=1, output_dir='output', visualization=False):
# 加载数据处理器
datas = self.load_datas(paths, images)
......
modules/image/semantic_segmentation/deeplabv3p_resnet50_voc/layers.py
浏览文件 @ d6b6c6ff
......@@ -30,8 +30,7 @@ def SyncBatchNorm(*args, **kwargs):
class ConvBNLayer(nn.Layer):
"""Basic conv bn relu layer."""
def __init__(
self,
def __init__(self,
in_channels: int,
out_channels: int,
kernel_size: int,
......@@ -44,8 +43,7 @@ class ConvBNLayer(nn.Layer):
super(ConvBNLayer, self).__init__()
self.is_vd_mode = is_vd_mode
self._pool2d_avg = AvgPool2D(
kernel_size=2, stride=2, padding=0, ceil_mode=True)
self._pool2d_avg = AvgPool2D(kernel_size=2, stride=2, padding=0, ceil_mode=True)
self._conv = Conv2D(
in_channels=in_channels,
out_channels=out_channels,
......@@ -83,11 +81,7 @@ class BottleneckBlock(nn.Layer):
super(BottleneckBlock, self).__init__()
self.conv0 = ConvBNLayer(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=1,
act='relu',
name=name + "_branch2a")
in_channels=in_channels, out_channels=out_channels, kernel_size=1, act='relu', name=name + "_branch2a")
self.dilation = dilation
......@@ -100,11 +94,7 @@ class BottleneckBlock(nn.Layer):
dilation=dilation,
name=name + "_branch2b")
self.conv2 = ConvBNLayer(
in_channels=out_channels,
out_channels=out_channels * 4,
kernel_size=1,
act=None,
name=name + "_branch2c")
in_channels=out_channels, out_channels=out_channels * 4, kernel_size=1, act=None, name=name + "_branch2c")
if not shortcut:
self.short = ConvBNLayer(
......@@ -139,12 +129,7 @@ class BottleneckBlock(nn.Layer):
class SeparableConvBNReLU(nn.Layer):
"""Depthwise Separable Convolution."""
def __init__(self,
in_channels: int,
out_channels: int,
kernel_size: int,
padding: str = 'same',
**kwargs: dict):
def __init__(self, in_channels: int, out_channels: int, kernel_size: int, padding: str = 'same', **kwargs: dict):
super(SeparableConvBNReLU, self).__init__()
self.depthwise_conv = ConvBN(
in_channels,
......@@ -153,8 +138,7 @@ class SeparableConvBNReLU(nn.Layer):
padding=padding,
groups=in_channels,
**kwargs)
self.piontwise_conv = ConvBNReLU(
in_channels, out_channels, kernel_size=1, groups=1)
self.piontwise_conv = ConvBNReLU(in_channels, out_channels, kernel_size=1, groups=1)
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
x = self.depthwise_conv(x)
......@@ -165,15 +149,9 @@ class SeparableConvBNReLU(nn.Layer):
class ConvBN(nn.Layer):
"""Basic conv bn layer"""
def __init__(self,
in_channels: int,
out_channels: int,
kernel_size: int,
padding: str = 'same',
**kwargs: dict):
def __init__(self, in_channels: int, out_channels: int, kernel_size: int, padding: str = 'same', **kwargs: dict):
super(ConvBN, self).__init__()
self._conv = Conv2D(
in_channels, out_channels, kernel_size, padding=padding, **kwargs)
self._conv = Conv2D(in_channels, out_channels, kernel_size, padding=padding, **kwargs)
self._batch_norm = SyncBatchNorm(out_channels)
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
......@@ -185,16 +163,10 @@ class ConvBN(nn.Layer):
class ConvBNReLU(nn.Layer):
"""Basic conv bn relu layer."""
def __init__(self,
in_channels: int,
out_channels: int,
kernel_size: int,
padding: str = 'same',
**kwargs: dict):
def __init__(self, in_channels: int, out_channels: int, kernel_size: int, padding: str = 'same', **kwargs: dict):
super(ConvBNReLU, self).__init__()
self._conv = Conv2D(
in_channels, out_channels, kernel_size, padding=padding, **kwargs)
self._conv = Conv2D(in_channels, out_channels, kernel_size, padding=padding, **kwargs)
self._batch_norm = SyncBatchNorm(out_channels)
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
......@@ -251,8 +223,7 @@ class Activation(nn.Layer):
act_name = act_dict[act]
self.act_func = eval("activation.{}()".format(act_name))
else:
raise KeyError("{} does not exist in the current {}".format(
act, act_dict.keys()))
raise KeyError("{} does not exist in the current {}".format(act, act_dict.keys()))
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
......@@ -281,7 +252,7 @@ class ASPPModule(nn.Layer):
in_channels: int,
out_channels: int,
align_corners: bool,
use_sep_conv: bool= False,
use_sep_conv: bool = False,
image_pooling: bool = False):
super().__init__()
......@@ -311,10 +282,7 @@ class ASPPModule(nn.Layer):
out_size += 1
self.image_pooling = image_pooling
self.conv_bn_relu = ConvBNReLU(
in_channels=out_channels * out_size,
out_channels=out_channels,
kernel_size=1)
self.conv_bn_relu = ConvBNReLU(in_channels=out_channels * out_size, out_channels=out_channels, kernel_size=1)
self.dropout = nn.Dropout(p=0.1) # drop rate
......@@ -322,20 +290,12 @@ class ASPPModule(nn.Layer):
outputs = []
for block in self.aspp_blocks:
y = block(x)
y = F.interpolate(
y,
x.shape[2:],
mode='bilinear',
align_corners=self.align_corners)
y = F.interpolate(y, x.shape[2:], mode='bilinear', align_corners=self.align_corners)
outputs.append(y)
if self.image_pooling:
img_avg = self.global_avg_pool(x)
img_avg = F.interpolate(
img_avg,
x.shape[2:],
mode='bilinear',
align_corners=self.align_corners)
img_avg = F.interpolate(img_avg, x.shape[2:], mode='bilinear', align_corners=self.align_corners)
outputs.append(img_avg)
x = paddle.concat(outputs, axis=1)
......
modules/image/semantic_segmentation/deeplabv3p_resnet50_voc/module.py
浏览文件 @ d6b6c6ff
......@@ -27,7 +27,6 @@ from deeplabv3p_resnet50_voc.resnet import ResNet50_vd
import deeplabv3p_resnet50_voc.layers as L
@moduleinfo(
name="deeplabv3p_resnet50_voc",
type="CV/semantic_segmentation",
......@@ -72,9 +71,8 @@ class DeepLabV3PResnet50(nn.Layer):
super(DeepLabV3PResnet50, self).__init__()
self.backbone = ResNet50_vd()
backbone_channels = [self.backbone.feat_channels[i] for i in backbone_indices]
self.head = DeepLabV3PHead(num_classes, backbone_indices,
backbone_channels, aspp_ratios,
aspp_out_channels, align_corners)
self.head = DeepLabV3PHead(num_classes, backbone_indices, backbone_channels, aspp_ratios, aspp_out_channels,
align_corners)
self.align_corners = align_corners
self.transforms = T.Compose([T.Normalize()])
......@@ -96,11 +94,8 @@ class DeepLabV3PResnet50(nn.Layer):
feat_list = self.backbone(x)
logit_list = self.head(feat_list)
return [
F.interpolate(
logit,
x.shape[2:],
mode='bilinear',
align_corners=self.align_corners) for logit in logit_list]
F.interpolate(logit, x.shape[2:], mode='bilinear', align_corners=self.align_corners) for logit in logit_list
]
class DeepLabV3PHead(nn.Layer):
......@@ -123,17 +118,13 @@ class DeepLabV3PHead(nn.Layer):
is even, e.g. 1024x512, otherwise it is True, e.g. 769x769.
"""
def __init__(self, num_classes: int, backbone_indices: Tuple[paddle.Tensor], backbone_channels: Tuple[paddle.Tensor],
aspp_ratios: Tuple[float], aspp_out_channels: int, align_corners: bool):
def __init__(self, num_classes: int, backbone_indices: Tuple[paddle.Tensor],
backbone_channels: Tuple[paddle.Tensor], aspp_ratios: Tuple[float], aspp_out_channels: int,
align_corners: bool):
super().__init__()
self.aspp = L.ASPPModule(
aspp_ratios,
backbone_channels[1],
aspp_out_channels,
align_corners,
use_sep_conv=True,
image_pooling=True)
aspp_ratios, backbone_channels[1], aspp_out_channels, align_corners, use_sep_conv=True, image_pooling=True)
self.decoder = Decoder(num_classes, backbone_channels[0], align_corners)
self.backbone_indices = backbone_indices
......@@ -160,25 +151,17 @@ class Decoder(nn.Layer):
def __init__(self, num_classes: int, in_channels: int, align_corners: bool):
super(Decoder, self).__init__()
self.conv_bn_relu1 = L.ConvBNReLU(
in_channels=in_channels, out_channels=48, kernel_size=1)
self.conv_bn_relu1 = L.ConvBNReLU(in_channels=in_channels, out_channels=48, kernel_size=1)
self.conv_bn_relu2 = L.SeparableConvBNReLU(
in_channels=304, out_channels=256, kernel_size=3, padding=1)
self.conv_bn_relu3 = L.SeparableConvBNReLU(
in_channels=256, out_channels=256, kernel_size=3, padding=1)
self.conv = nn.Conv2D(
in_channels=256, out_channels=num_classes, kernel_size=1)
self.conv_bn_relu2 = L.SeparableConvBNReLU(in_channels=304, out_channels=256, kernel_size=3, padding=1)
self.conv_bn_relu3 = L.SeparableConvBNReLU(in_channels=256, out_channels=256, kernel_size=3, padding=1)
self.conv = nn.Conv2D(in_channels=256, out_channels=num_classes, kernel_size=1)
self.align_corners = align_corners
def forward(self, x: paddle.Tensor, low_level_feat: paddle.Tensor) -> paddle.Tensor:
low_level_feat = self.conv_bn_relu1(low_level_feat)
x = F.interpolate(
x,
low_level_feat.shape[2:],
mode='bilinear',
align_corners=self.align_corners)
x = F.interpolate(x, low_level_feat.shape[2:], mode='bilinear', align_corners=self.align_corners)
x = paddle.concat([x, low_level_feat], axis=1)
x = self.conv_bn_relu2(x)
x = self.conv_bn_relu3(x)
......
modules/image/semantic_segmentation/deeplabv3p_resnet50_voc/resnet.py
浏览文件 @ d6b6c6ff
......@@ -36,11 +36,7 @@ class BasicBlock(nn.Layer):
act='relu',
name=name + "_branch2a")
self.conv1 = L.ConvBNLayer(
in_channels=out_channels,
out_channels=out_channels,
kernel_size=3,
act=None,
name=name + "_branch2b")
in_channels=out_channels, out_channels=out_channels, kernel_size=3, act=None, name=name + "_branch2b")
if not shortcut:
self.short = L.ConvBNLayer(
......@@ -67,8 +63,7 @@ class BasicBlock(nn.Layer):
class ResNet50_vd(nn.Layer):
def __init__(self,
multi_grid: tuple = (1, 2, 4)):
def __init__(self, multi_grid: tuple = (1, 2, 4)):
super(ResNet50_vd, self).__init__()
depth = [3, 4, 6, 3]
num_channels = [64, 256, 512, 1024]
......@@ -76,26 +71,11 @@ class ResNet50_vd(nn.Layer):
self.feat_channels = [c * 4 for c in num_filters]
dilation_dict = {2: 2, 3: 4}
self.conv1_1 = L.ConvBNLayer(
in_channels=3,
out_channels=32,
kernel_size=3,
stride=2,
act='relu',
name="conv1_1")
in_channels=3, out_channels=32, kernel_size=3, stride=2, act='relu', name="conv1_1")
self.conv1_2 = L.ConvBNLayer(
in_channels=32,
out_channels=32,
kernel_size=3,
stride=1,
act='relu',
name="conv1_2")
in_channels=32, out_channels=32, kernel_size=3, stride=1, act='relu', name="conv1_2")
self.conv1_3 = L.ConvBNLayer(
in_channels=32,
out_channels=64,
kernel_size=3,
stride=1,
act='relu',
name="conv1_3")
in_channels=32, out_channels=64, kernel_size=3, stride=1, act='relu', name="conv1_3")
self.pool2d_max = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
self.stage_list = []
......@@ -104,18 +84,15 @@ class ResNet50_vd(nn.Layer):
block_list = []
for i in range(depth[block]):
conv_name = "res" + str(block + 2) + chr(97 + i)
dilation_rate = dilation_dict[
block] if dilation_dict and block in dilation_dict else 1
dilation_rate = dilation_dict[block] if dilation_dict and block in dilation_dict else 1
if block == 3:
dilation_rate = dilation_rate * multi_grid[i]
bottleneck_block = self.add_sublayer(
'bb_%d_%d' % (block, i),
L.BottleneckBlock(
in_channels=num_channels[block]
if i == 0 else num_filters[block] * 4,
in_channels=num_channels[block] if i == 0 else num_filters[block] * 4,
out_channels=num_filters[block],
stride=2 if i == 0 and block != 0
and dilation_rate == 1 else 1,
stride=2 if i == 0 and block != 0 and dilation_rate == 1 else 1,
shortcut=shortcut,
if_first=block == i == 0,
name=conv_name,
......
modules/image/semantic_segmentation/ocrnet_hrnetw18_voc/hrnet.py
浏览文件 @ d6b6c6ff
......@@ -51,8 +51,8 @@ class HRNet_W18(nn.Layer):
def __init__(self,
pretrained: str = None,
stage1_num_modules: int = 1,
stage1_num_blocks: tuple = (4,),
stage1_num_channels: tuple = (64,),
stage1_num_blocks: tuple = (4, ),
stage1_num_channels: tuple = (64, ),
stage2_num_modules: int = 1,
stage2_num_blocks: tuple = (4, 4),
stage2_num_channels: tuple = (18, 36),
......@@ -83,20 +83,10 @@ class HRNet_W18(nn.Layer):
self.feat_channels = [sum(stage4_num_channels)]
self.conv_layer1_1 = L.ConvBNReLU(
in_channels=3,
out_channels=64,
kernel_size=3,
stride=2,
padding='same',
bias_attr=False)
in_channels=3, out_channels=64, kernel_size=3, stride=2, padding='same', bias_attr=False)
self.conv_layer1_2 = L.ConvBNReLU(
in_channels=64,
out_channels=64,
kernel_size=3,
stride=2,
padding='same',
bias_attr=False)
in_channels=64, out_channels=64, kernel_size=3, stride=2, padding='same', bias_attr=False)
self.la1 = Layer1(
num_channels=64,
......@@ -106,9 +96,7 @@ class HRNet_W18(nn.Layer):
name="layer2")
self.tr1 = TransitionLayer(
in_channels=[self.stage1_num_channels[0] * 4],
out_channels=self.stage2_num_channels,
name="tr1")
in_channels=[self.stage1_num_channels[0] * 4], out_channels=self.stage2_num_channels, name="tr1")
self.st2 = Stage(
num_channels=self.stage2_num_channels,
......@@ -120,9 +108,7 @@ class HRNet_W18(nn.Layer):
align_corners=align_corners)
self.tr2 = TransitionLayer(
in_channels=self.stage2_num_channels,
out_channels=self.stage3_num_channels,
name="tr2")
in_channels=self.stage2_num_channels, out_channels=self.stage3_num_channels, name="tr2")
self.st3 = Stage(
num_channels=self.stage3_num_channels,
num_modules=self.stage3_num_modules,
......@@ -133,9 +119,7 @@ class HRNet_W18(nn.Layer):
align_corners=align_corners)
self.tr3 = TransitionLayer(
in_channels=self.stage3_num_channels,
out_channels=self.stage4_num_channels,
name="tr3")
in_channels=self.stage3_num_channels, out_channels=self.stage4_num_channels, name="tr3")
self.st4 = Stage(
num_channels=self.stage4_num_channels,
num_modules=self.stage4_num_modules,
......@@ -161,30 +145,16 @@ class HRNet_W18(nn.Layer):
st4 = self.st4(tr3)
x0_h, x0_w = st4[0].shape[2:]
x1 = F.interpolate(
st4[1], (x0_h, x0_w),
mode='bilinear',
align_corners=self.align_corners)
x2 = F.interpolate(
st4[2], (x0_h, x0_w),
mode='bilinear',
align_corners=self.align_corners)
x3 = F.interpolate(
st4[3], (x0_h, x0_w),
mode='bilinear',
align_corners=self.align_corners)
x1 = F.interpolate(st4[1], (x0_h, x0_w), mode='bilinear', align_corners=self.align_corners)
x2 = F.interpolate(st4[2], (x0_h, x0_w), mode='bilinear', align_corners=self.align_corners)
x3 = F.interpolate(st4[3], (x0_h, x0_w), mode='bilinear', align_corners=self.align_corners)
x = paddle.concat([st4[0], x1, x2, x3], axis=1)
return [x]
class Layer1(nn.Layer):
def __init__(self,
num_channels: int,
num_filters: int,
num_blocks: int,
has_se: bool = False,
name: str = None):
def __init__(self, num_channels: int, num_filters: int, num_blocks: int, has_se: bool = False, name: str = None):
super(Layer1, self).__init__()
self.bottleneck_block_list = []
......@@ -253,12 +223,7 @@ class TransitionLayer(nn.Layer):
class Branches(nn.Layer):
def __init__(self,
num_blocks: int,
in_channels: int,
out_channels: int,
has_se: bool = False,
name: str = None):
def __init__(self, num_blocks: int, in_channels: int, out_channels: int, has_se: bool = False, name: str = None):
super(Branches, self).__init__()
self.basic_block_list = []
......@@ -273,8 +238,7 @@ class Branches(nn.Layer):
num_channels=in_ch,
num_filters=out_channels[i],
has_se=has_se,
name=name + '_branch_layer_' + str(i + 1) + '_' +
str(j + 1)))
name=name + '_branch_layer_' + str(i + 1) + '_' + str(j + 1)))
self.basic_block_list[i].append(basic_block_func)
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
......@@ -301,11 +265,7 @@ class BottleneckBlock(nn.Layer):
self.downsample = downsample
self.conv1 = L.ConvBNReLU(
in_channels=num_channels,
out_channels=num_filters,
kernel_size=1,
padding='same',
bias_attr=False)
in_channels=num_channels, out_channels=num_filters, kernel_size=1, padding='same', bias_attr=False)
self.conv2 = L.ConvBNReLU(
in_channels=num_filters,
......@@ -316,26 +276,15 @@ class BottleneckBlock(nn.Layer):
bias_attr=False)
self.conv3 = L.ConvBN(
in_channels=num_filters,
out_channels=num_filters * 4,
kernel_size=1,
padding='same',
bias_attr=False)
in_channels=num_filters, out_channels=num_filters * 4, kernel_size=1, padding='same', bias_attr=False)
if self.downsample:
self.conv_down = L.ConvBN(
in_channels=num_channels,
out_channels=num_filters * 4,
kernel_size=1,
padding='same',
bias_attr=False)
in_channels=num_channels, out_channels=num_filters * 4, kernel_size=1, padding='same', bias_attr=False)
if self.has_se:
self.se = SELayer(
num_channels=num_filters * 4,
num_filters=num_filters * 4,
reduction_ratio=16,
name=name + '_fc')
num_channels=num_filters * 4, num_filters=num_filters * 4, reduction_ratio=16, name=name + '_fc')
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
residual = x
......@@ -375,26 +324,14 @@ class BasicBlock(nn.Layer):
padding='same',
bias_attr=False)
self.conv2 = L.ConvBN(
in_channels=num_filters,
out_channels=num_filters,
kernel_size=3,
padding='same',
bias_attr=False)
in_channels=num_filters, out_channels=num_filters, kernel_size=3, padding='same', bias_attr=False)
if self.downsample:
self.conv_down = L.ConvBNReLU(
in_channels=num_channels,
out_channels=num_filters,
kernel_size=1,
padding='same',
bias_attr=False)
in_channels=num_channels, out_channels=num_filters, kernel_size=1, padding='same', bias_attr=False)
if self.has_se:
self.se = SELayer(
num_channels=num_filters,
num_filters=num_filters,
reduction_ratio=16,
name=name + '_fc')
self.se = SELayer(num_channels=num_filters, num_filters=num_filters, reduction_ratio=16, name=name + '_fc')
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
residual = x
......@@ -423,17 +360,11 @@ class SELayer(nn.Layer):
med_ch = int(num_channels / reduction_ratio)
stdv = 1.0 / math.sqrt(num_channels * 1.0)
self.squeeze = nn.Linear(
num_channels,
med_ch,
weight_attr=paddle.ParamAttr(
initializer=nn.initializer.Uniform(-stdv, stdv)))
num_channels, med_ch, weight_attr=paddle.ParamAttr(initializer=nn.initializer.Uniform(-stdv, stdv)))
stdv = 1.0 / math.sqrt(med_ch * 1.0)
self.excitation = nn.Linear(
med_ch,
num_filters,
weight_attr=paddle.ParamAttr(
initializer=nn.initializer.Uniform(-stdv, stdv)))
med_ch, num_filters, weight_attr=paddle.ParamAttr(initializer=nn.initializer.Uniform(-stdv, stdv)))
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
pool = self.pool2d_gap(x)
......@@ -442,8 +373,7 @@ class SELayer(nn.Layer):
squeeze = F.relu(squeeze)
excitation = self.excitation(squeeze)
excitation = F.sigmoid(excitation)
excitation = paddle.reshape(
excitation, shape=[-1, self._num_channels, 1, 1])
excitation = paddle.reshape(excitation, shape=[-1, self._num_channels, 1, 1])
out = x * excitation
return out
......@@ -507,11 +437,7 @@ class HighResolutionModule(nn.Layer):
super(HighResolutionModule, self).__init__()
self.branches_func = Branches(
num_blocks=num_blocks,
in_channels=num_channels,
out_channels=num_filters,
has_se=has_se,
name=name)
num_blocks=num_blocks, in_channels=num_channels, out_channels=num_filters, has_se=has_se, name=name)
self.fuse_func = FuseLayers(
in_channels=num_filters,
......@@ -557,8 +483,7 @@ class FuseLayers(nn.Layer):
for k in range(i - j):
if k == i - j - 1:
residual_func = self.add_sublayer(
"residual_{}_layer_{}_{}_{}".format(
name, i + 1, j + 1, k + 1),
"residual_{}_layer_{}_{}_{}".format(name, i + 1, j + 1, k + 1),
L.ConvBN(
in_channels=pre_num_filters,
out_channels=out_channels[i],
......@@ -569,8 +494,7 @@ class FuseLayers(nn.Layer):
pre_num_filters = out_channels[i]
else:
residual_func = self.add_sublayer(
"residual_{}_layer_{}_{}_{}".format(
name, i + 1, j + 1, k + 1),
"residual_{}_layer_{}_{}_{}".format(name, i + 1, j + 1, k + 1),
L.ConvBNReLU(
in_channels=pre_num_filters,
out_channels=out_channels[j],
......@@ -592,11 +516,7 @@ class FuseLayers(nn.Layer):
y = self.residual_func_list[residual_func_idx](x[j])
residual_func_idx += 1
y = F.interpolate(
y,
residual_shape,
mode='bilinear',
align_corners=self.align_corners)
y = F.interpolate(y, residual_shape, mode='bilinear', align_corners=self.align_corners)
residual = residual + y
elif j < i:
y = x[j]
......
modules/image/semantic_segmentation/ocrnet_hrnetw18_voc/layers.py
浏览文件 @ d6b6c6ff
......@@ -30,8 +30,7 @@ def SyncBatchNorm(*args, **kwargs):
class ConvBNLayer(nn.Layer):
"""Basic conv bn relu layer."""
def __init__(
self,
def __init__(self,
in_channels: int,
out_channels: int,
kernel_size: int,
......@@ -44,8 +43,7 @@ class ConvBNLayer(nn.Layer):
super(ConvBNLayer, self).__init__()
self.is_vd_mode = is_vd_mode
self._pool2d_avg = AvgPool2D(
kernel_size=2, stride=2, padding=0, ceil_mode=True)
self._pool2d_avg = AvgPool2D(kernel_size=2, stride=2, padding=0, ceil_mode=True)
self._conv = Conv2D(
in_channels=in_channels,
out_channels=out_channels,
......@@ -83,11 +81,7 @@ class BottleneckBlock(nn.Layer):
super(BottleneckBlock, self).__init__()
self.conv0 = ConvBNLayer(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=1,
act='relu',
name=name + "_branch2a")
in_channels=in_channels, out_channels=out_channels, kernel_size=1, act='relu', name=name + "_branch2a")
self.dilation = dilation
......@@ -100,11 +94,7 @@ class BottleneckBlock(nn.Layer):
dilation=dilation,
name=name + "_branch2b")
self.conv2 = ConvBNLayer(
in_channels=out_channels,
out_channels=out_channels * 4,
kernel_size=1,
act=None,
name=name + "_branch2c")
in_channels=out_channels, out_channels=out_channels * 4, kernel_size=1, act=None, name=name + "_branch2c")
if not shortcut:
self.short = ConvBNLayer(
......@@ -139,12 +129,7 @@ class BottleneckBlock(nn.Layer):
class SeparableConvBNReLU(nn.Layer):
"""Depthwise Separable Convolution."""
def __init__(self,
in_channels: int,
out_channels: int,
kernel_size: int,
padding: str = 'same',
**kwargs: dict):
def __init__(self, in_channels: int, out_channels: int, kernel_size: int, padding: str = 'same', **kwargs: dict):
super(SeparableConvBNReLU, self).__init__()
self.depthwise_conv = ConvBN(
in_channels,
......@@ -153,8 +138,7 @@ class SeparableConvBNReLU(nn.Layer):
padding=padding,
groups=in_channels,
**kwargs)
self.piontwise_conv = ConvBNReLU(
in_channels, out_channels, kernel_size=1, groups=1)
self.piontwise_conv = ConvBNReLU(in_channels, out_channels, kernel_size=1, groups=1)
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
x = self.depthwise_conv(x)
......@@ -165,15 +149,9 @@ class SeparableConvBNReLU(nn.Layer):
class ConvBN(nn.Layer):
"""Basic conv bn layer"""
def __init__(self,
in_channels: int,
out_channels: int,
kernel_size: int,
padding: str = 'same',
**kwargs: dict):
def __init__(self, in_channels: int, out_channels: int, kernel_size: int, padding: str = 'same', **kwargs: dict):
super(ConvBN, self).__init__()
self._conv = Conv2D(
in_channels, out_channels, kernel_size, padding=padding, **kwargs)
self._conv = Conv2D(in_channels, out_channels, kernel_size, padding=padding, **kwargs)
self._batch_norm = SyncBatchNorm(out_channels)
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
......@@ -185,16 +163,10 @@ class ConvBN(nn.Layer):
class ConvBNReLU(nn.Layer):
"""Basic conv bn relu layer."""
def __init__(self,
in_channels: int,
out_channels: int,
kernel_size: int,
padding: str = 'same',
**kwargs: dict):
def __init__(self, in_channels: int, out_channels: int, kernel_size: int, padding: str = 'same', **kwargs: dict):
super(ConvBNReLU, self).__init__()
self._conv = Conv2D(
in_channels, out_channels, kernel_size, padding=padding, **kwargs)
self._conv = Conv2D(in_channels, out_channels, kernel_size, padding=padding, **kwargs)
self._batch_norm = SyncBatchNorm(out_channels)
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
......@@ -251,8 +223,7 @@ class Activation(nn.Layer):
act_name = act_dict[act]
self.act_func = eval("activation.{}()".format(act_name))
else:
raise KeyError("{} does not exist in the current {}".format(
act, act_dict.keys()))
raise KeyError("{} does not exist in the current {}".format(act, act_dict.keys()))
def forward(self, x: paddle.Tensor) -> paddle.Tensor:
......@@ -276,13 +247,7 @@ class ASPPModule(nn.Layer):
image_pooling (bool, optional): If augmented with image-level features. Default: False
"""
def __init__(self,
aspp_ratios,
in_channels,
out_channels,
align_corners,
use_sep_conv=False,
image_pooling=False):
def __init__(self, aspp_ratios, in_channels, out_channels, align_corners, use_sep_conv=False, image_pooling=False):
super().__init__()
self.align_corners = align_corners
......@@ -311,10 +276,7 @@ class ASPPModule(nn.Layer):
out_size += 1
self.image_pooling = image_pooling
self.conv_bn_relu = ConvBNReLU(
in_channels=out_channels * out_size,
out_channels=out_channels,
kernel_size=1)
self.conv_bn_relu = ConvBNReLU(in_channels=out_channels * out_size, out_channels=out_channels, kernel_size=1)
self.dropout = nn.Dropout(p=0.1) # drop rate
......@@ -322,20 +284,12 @@ class ASPPModule(nn.Layer):
outputs = []
for block in self.aspp_blocks:
y = block(x)
y = F.interpolate(
y,
x.shape[2:],
mode='bilinear',
align_corners=self.align_corners)
y = F.interpolate(y, x.shape[2:], mode='bilinear', align_corners=self.align_corners)
outputs.append(y)
if self.image_pooling:
img_avg = self.global_avg_pool(x)
img_avg = F.interpolate(
img_avg,
x.shape[2:],
mode='bilinear',
align_corners=self.align_corners)
img_avg = F.interpolate(img_avg, x.shape[2:], mode='bilinear', align_corners=self.align_corners)
outputs.append(img_avg)
x = paddle.concat(outputs, axis=1)
......
modules/image/semantic_segmentation/ocrnet_hrnetw18_voc/module.py
浏览文件 @ d6b6c6ff
......@@ -25,6 +25,7 @@ from paddlehub.module.cv_module import ImageSegmentationModule
import ocrnet_hrnetw18_voc.layers as L
from ocrnet_hrnetw18_voc.hrnet import HRNet_W18
@moduleinfo(
name="ocrnet_hrnetw18_voc",
type="CV/semantic_segmentation",
......@@ -90,11 +91,8 @@ class OCRNetHRNetW18(nn.Layer):
feats = [feats[i] for i in self.backbone_indices]
logit_list = self.head(feats)
logit_list = [
F.interpolate(
logit,
x.shape[2:],
mode='bilinear',
align_corners=self.align_corners) for logit in logit_list]
F.interpolate(logit, x.shape[2:], mode='bilinear', align_corners=self.align_corners) for logit in logit_list
]
return logit_list
......@@ -108,32 +106,23 @@ class OCRHead(nn.Layer):
ocr_key_channels(int, optional): The number of key channels in ObjectAttentionBlock. Default: 256.
"""
def __init__(self,
num_classes: int,
in_channels: int,
ocr_mid_channels: int = 512,
ocr_key_channels: int = 256):
def __init__(self, num_classes: int, in_channels: int, ocr_mid_channels: int = 512, ocr_key_channels: int = 256):
super().__init__()
self.num_classes = num_classes
self.spatial_gather = SpatialGatherBlock()
self.spatial_ocr = SpatialOCRModule(ocr_mid_channels, ocr_key_channels,
ocr_mid_channels)
self.spatial_ocr = SpatialOCRModule(ocr_mid_channels, ocr_key_channels, ocr_mid_channels)
self.indices = [-2, -1] if len(in_channels) > 1 else [-1, -1]
self.conv3x3_ocr = L.ConvBNReLU(
in_channels[self.indices[1]], ocr_mid_channels, 3, padding=1)
self.conv3x3_ocr = L.ConvBNReLU(in_channels[self.indices[1]], ocr_mid_channels, 3, padding=1)
self.cls_head = nn.Conv2D(ocr_mid_channels, self.num_classes, 1)
self.aux_head = nn.Sequential(
L.ConvBNReLU(in_channels[self.indices[0]],
in_channels[self.indices[0]], 1),
L.ConvBNReLU(in_channels[self.indices[0]], in_channels[self.indices[0]], 1),
nn.Conv2D(in_channels[self.indices[0]], self.num_classes, 1))
def forward(self, feat_list: List[paddle.Tensor]) -> paddle.Tensor:
feat_shallow, feat_deep = feat_list[self.indices[0]], feat_list[
self.indices[1]]
feat_shallow, feat_deep = feat_list[self.indices[0]], feat_list[self.indices[1]]
soft_regions = self.aux_head(feat_shallow)
pixels = self.conv3x3_ocr(feat_deep)
......@@ -171,17 +160,11 @@ class SpatialGatherBlock(nn.Layer):
class SpatialOCRModule(nn.Layer):
"""Aggregate the global object representation to update the representation for each pixel."""
def __init__(self,
in_channels: int,
key_channels: int,
out_channels: int,
dropout_rate: float = 0.1):
def __init__(self, in_channels: int, key_channels: int, out_channels: int, dropout_rate: float = 0.1):
super().__init__()
self.attention_block = ObjectAttentionBlock(in_channels, key_channels)
self.conv1x1 = nn.Sequential(
L.ConvBNReLU(2 * in_channels, out_channels, 1),
nn.Dropout2D(dropout_rate))
self.conv1x1 = nn.Sequential(L.ConvBNReLU(2 * in_channels, out_channels, 1), nn.Dropout2D(dropout_rate))
def forward(self, pixels: paddle.Tensor, regions: paddle.Tensor) -> paddle.Tensor:
context = self.attention_block(pixels, regions)
......@@ -201,12 +184,10 @@ class ObjectAttentionBlock(nn.Layer):
self.key_channels = key_channels
self.f_pixel = nn.Sequential(
L.ConvBNReLU(in_channels, key_channels, 1),
L.ConvBNReLU(key_channels, key_channels, 1))
L.ConvBNReLU(in_channels, key_channels, 1), L.ConvBNReLU(key_channels, key_channels, 1))
self.f_object = nn.Sequential(
L.ConvBNReLU(in_channels, key_channels, 1),
L.ConvBNReLU(key_channels, key_channels, 1))
L.ConvBNReLU(in_channels, key_channels, 1), L.ConvBNReLU(key_channels, key_channels, 1))
self.f_down = L.ConvBNReLU(in_channels, key_channels, 1)
......
modules/text/language_model/bert-base-cased/module.py
浏览文件 @ d6b6c6ff
......@@ -60,29 +60,27 @@ class Bert(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = BertForSequenceClassification.from_pretrained(pretrained_model_name_or_path='bert-base-cased', num_classes=self.num_classes, **kwargs)
self.model = BertForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='bert-base-cased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = BertForTokenClassification.from_pretrained(pretrained_model_name_or_path='bert-base-cased', num_classes=self.num_classes, **kwargs)
self.model = BertForTokenClassification.from_pretrained(
pretrained_model_name_or_path='bert-base-cased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-base-cased', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-base-cased', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -130,8 +128,7 @@ class Bert(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -173,5 +170,4 @@ class Bert(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return BertTokenizer.from_pretrained(
pretrained_model_name_or_path='bert-base-cased', *args, **kwargs)
return BertTokenizer.from_pretrained(pretrained_model_name_or_path='bert-base-cased', *args, **kwargs)
modules/text/language_model/bert-base-chinese/module.py
浏览文件 @ d6b6c6ff
......@@ -60,37 +60,27 @@ class Bert(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = BertForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='bert-base-chinese',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='bert-base-chinese', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = BertForTokenClassification.from_pretrained(
pretrained_model_name_or_path='bert-base-chinese',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='bert-base-chinese', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-base-chinese', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-base-chinese', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +128,7 @@ class Bert(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -181,5 +170,4 @@ class Bert(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return BertTokenizer.from_pretrained(
pretrained_model_name_or_path='bert-base-chinese', *args, **kwargs)
return BertTokenizer.from_pretrained(pretrained_model_name_or_path='bert-base-chinese', *args, **kwargs)
modules/text/language_model/bert-base-multilingual-cased/module.py
浏览文件 @ d6b6c6ff
......@@ -60,37 +60,29 @@ class Bert(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = BertForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='bert-base-multilingual-cased',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='bert-base-multilingual-cased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = BertForTokenClassification.from_pretrained(
pretrained_model_name_or_path='bert-base-multilingual-cased',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='bert-base-multilingual-cased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-base-multilingual-cased', **kwargs)
self.model = BertModel.from_pretrained(
pretrained_model_name_or_path='bert-base-multilingual-cased', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-base-multilingual-cased', **kwargs)
self.model = BertModel.from_pretrained(
pretrained_model_name_or_path='bert-base-multilingual-cased', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +130,7 @@ class Bert(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......
modules/text/language_model/bert-base-multilingual-uncased/module.py
浏览文件 @ d6b6c6ff
......@@ -60,37 +60,29 @@ class Bert(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = BertForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='bert-base-multilingual-uncased',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='bert-base-multilingual-uncased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = BertForTokenClassification.from_pretrained(
pretrained_model_name_or_path='bert-base-multilingual-uncased',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='bert-base-multilingual-uncased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-base-multilingual-uncased', **kwargs)
self.model = BertModel.from_pretrained(
pretrained_model_name_or_path='bert-base-multilingual-uncased', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-base-multilingual-uncased', **kwargs)
self.model = BertModel.from_pretrained(
pretrained_model_name_or_path='bert-base-multilingual-uncased', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +130,7 @@ class Bert(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......
modules/text/language_model/bert-base-uncased/module.py
浏览文件 @ d6b6c6ff
......@@ -60,29 +60,27 @@ class Bert(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = BertForSequenceClassification.from_pretrained(pretrained_model_name_or_path='bert-base-uncased', num_classes=self.num_classes, **kwargs)
self.model = BertForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='bert-base-uncased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = BertForTokenClassification.from_pretrained(pretrained_model_name_or_path='bert-base-uncased', num_classes=self.num_classes, **kwargs)
self.model = BertForTokenClassification.from_pretrained(
pretrained_model_name_or_path='bert-base-uncased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-base-uncased', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-base-uncased', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -130,8 +128,7 @@ class Bert(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -173,5 +170,4 @@ class Bert(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return BertTokenizer.from_pretrained(
pretrained_model_name_or_path='bert-base-uncased', *args, **kwargs)
return BertTokenizer.from_pretrained(pretrained_model_name_or_path='bert-base-uncased', *args, **kwargs)
modules/text/language_model/bert-large-cased/module.py
浏览文件 @ d6b6c6ff
......@@ -60,29 +60,27 @@ class Bert(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = BertForSequenceClassification.from_pretrained(pretrained_model_name_or_path='bert-large-cased', num_classes=self.num_classes, **kwargs)
self.model = BertForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='bert-large-cased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = BertForTokenClassification.from_pretrained(pretrained_model_name_or_path='bert-large-cased', num_classes=self.num_classes, **kwargs)
self.model = BertForTokenClassification.from_pretrained(
pretrained_model_name_or_path='bert-large-cased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-large-cased', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-large-cased', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -130,8 +128,7 @@ class Bert(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -173,5 +170,4 @@ class Bert(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return BertTokenizer.from_pretrained(
pretrained_model_name_or_path='bert-large-cased', *args, **kwargs)
return BertTokenizer.from_pretrained(pretrained_model_name_or_path='bert-large-cased', *args, **kwargs)
modules/text/language_model/bert-large-uncased/module.py
浏览文件 @ d6b6c6ff
......@@ -60,29 +60,27 @@ class Bert(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = BertForSequenceClassification.from_pretrained(pretrained_model_name_or_path='bert-large-uncased', num_classes=self.num_classes, **kwargs)
self.model = BertForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='bert-large-uncased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = BertForTokenClassification.from_pretrained(pretrained_model_name_or_path='bert-large-uncased', num_classes=self.num_classes, **kwargs)
self.model = BertForTokenClassification.from_pretrained(
pretrained_model_name_or_path='bert-large-uncased', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-large-uncased', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-large-uncased', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -130,8 +128,7 @@ class Bert(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -173,5 +170,4 @@ class Bert(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return BertTokenizer.from_pretrained(
pretrained_model_name_or_path='bert-large-uncased', *args, **kwargs)
return BertTokenizer.from_pretrained(pretrained_model_name_or_path='bert-large-uncased', *args, **kwargs)
modules/text/language_model/chinese_bert_wwm/module.py
浏览文件 @ d6b6c6ff
......@@ -35,8 +35,7 @@ from paddlehub.utils.log import logger
author="ymcui",
author_email="ymcui@ir.hit.edu.cn",
type="nlp/semantic_model",
meta=TransformerModule
)
meta=TransformerModule)
class BertWwm(nn.Layer):
"""
BertWwm model
......@@ -61,37 +60,27 @@ class BertWwm(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = BertForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='bert-wwm-chinese',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='bert-wwm-chinese', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = BertForTokenClassification.from_pretrained(
pretrained_model_name_or_path='bert-wwm-chinese',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='bert-wwm-chinese', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-wwm-chinese', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-wwm-chinese', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -139,8 +128,7 @@ class BertWwm(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -182,5 +170,4 @@ class BertWwm(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return BertTokenizer.from_pretrained(
pretrained_model_name_or_path='bert-wwm-chinese', *args, **kwargs)
return BertTokenizer.from_pretrained(pretrained_model_name_or_path='bert-wwm-chinese', *args, **kwargs)
modules/text/language_model/chinese_bert_wwm_ext/module.py
浏览文件 @ d6b6c6ff
......@@ -35,8 +35,7 @@ from paddlehub.utils.log import logger
author="ymcui",
author_email="ymcui@ir.hit.edu.cn",
type="nlp/semantic_model",
meta=TransformerModule
)
meta=TransformerModule)
class BertWwm(nn.Layer):
"""
BertWwm model
......@@ -61,37 +60,27 @@ class BertWwm(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = BertForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='bert-wwm-ext-chinese',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='bert-wwm-ext-chinese', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = BertForTokenClassification.from_pretrained(
pretrained_model_name_or_path='bert-wwm-ext-chinese',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='bert-wwm-ext-chinese', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-wwm-ext-chinese', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = BertModel.from_pretrained(pretrained_model_name_or_path='bert-wwm-ext-chinese', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -139,8 +128,7 @@ class BertWwm(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -182,5 +170,4 @@ class BertWwm(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return BertTokenizer.from_pretrained(
pretrained_model_name_or_path='bert-wwm-ext-chinese', *args, **kwargs)
return BertTokenizer.from_pretrained(pretrained_model_name_or_path='bert-wwm-ext-chinese', *args, **kwargs)
modules/text/language_model/chinese_electra_base/module.py
浏览文件 @ d6b6c6ff
......@@ -60,37 +60,27 @@ class Electra(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = ElectraForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='chinese-electra-base',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='chinese-electra-base', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = ElectraForTokenClassification.from_pretrained(
pretrained_model_name_or_path='chinese-electra-base',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='chinese-electra-base', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = ElectraModel.from_pretrained(pretrained_model_name_or_path='chinese-electra-base', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = ElectraModel.from_pretrained(pretrained_model_name_or_path='chinese-electra-base', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +128,7 @@ class Electra(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -181,5 +170,4 @@ class Electra(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return ElectraTokenizer.from_pretrained(
pretrained_model_name_or_path='chinese-electra-base', *args, **kwargs)
return ElectraTokenizer.from_pretrained(pretrained_model_name_or_path='chinese-electra-base', *args, **kwargs)
modules/text/language_model/chinese_electra_small/module.py
浏览文件 @ d6b6c6ff
......@@ -60,37 +60,27 @@ class Electra(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = ElectraForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='chinese-electra-small',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='chinese-electra-small', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = ElectraForTokenClassification.from_pretrained(
pretrained_model_name_or_path='chinese-electra-small',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='chinese-electra-small', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = ElectraModel.from_pretrained(pretrained_model_name_or_path='chinese-electra-small', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = ElectraModel.from_pretrained(pretrained_model_name_or_path='chinese-electra-small', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +128,7 @@ class Electra(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -181,5 +170,4 @@ class Electra(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return ElectraTokenizer.from_pretrained(
pretrained_model_name_or_path='chinese-electra-small', *args, **kwargs)
return ElectraTokenizer.from_pretrained(pretrained_model_name_or_path='chinese-electra-small', *args, **kwargs)
modules/text/language_model/electra_base/module.py
浏览文件 @ d6b6c6ff
......@@ -29,8 +29,7 @@ from paddlehub.utils.log import logger
@moduleinfo(
name="electra-base",
version="1.0.1",
summary=
"electra-base, 12-layer, 768-hidden, 12-heads, 110M parameters. The module is executed as paddle.dygraph.",
summary="electra-base, 12-layer, 768-hidden, 12-heads, 110M parameters. The module is executed as paddle.dygraph.",
author="paddlepaddle",
author_email="",
type="nlp/semantic_model",
......@@ -60,37 +59,27 @@ class Electra(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = ElectraForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='electra-base',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='electra-base', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = ElectraForTokenClassification.from_pretrained(
pretrained_model_name_or_path='electra-base',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='electra-base', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = ElectraModel.from_pretrained(pretrained_model_name_or_path='electra-base', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = ElectraModel.from_pretrained(pretrained_model_name_or_path='electra-base', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +127,7 @@ class Electra(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -181,5 +169,4 @@ class Electra(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return ElectraTokenizer.from_pretrained(
pretrained_model_name_or_path='electra-base', *args, **kwargs)
return ElectraTokenizer.from_pretrained(pretrained_model_name_or_path='electra-base', *args, **kwargs)
modules/text/language_model/electra_large/module.py
浏览文件 @ d6b6c6ff
......@@ -29,8 +29,7 @@ from paddlehub.utils.log import logger
@moduleinfo(
name="electra-large",
version="1.0.1",
summary=
"electra-large, 24-layer, 1024-hidden, 16-heads, 335M parameters. The module is executed as paddle.dygraph.",
summary="electra-large, 24-layer, 1024-hidden, 16-heads, 335M parameters. The module is executed as paddle.dygraph.",
author="paddlepaddle",
author_email="",
type="nlp/semantic_model",
......@@ -60,37 +59,27 @@ class Electra(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = ElectraForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='electra-large',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='electra-large', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = ElectraForTokenClassification.from_pretrained(
pretrained_model_name_or_path='electra-large',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='electra-large', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = ElectraModel.from_pretrained(pretrained_model_name_or_path='electra-large', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = ElectraModel.from_pretrained(pretrained_model_name_or_path='electra-large', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +127,7 @@ class Electra(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -181,5 +169,4 @@ class Electra(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return ElectraTokenizer.from_pretrained(
pretrained_model_name_or_path='electra-large', *args, **kwargs)
return ElectraTokenizer.from_pretrained(pretrained_model_name_or_path='electra-large', *args, **kwargs)
modules/text/language_model/electra_small/module.py
浏览文件 @ d6b6c6ff
......@@ -29,8 +29,7 @@ from paddlehub.utils.log import logger
@moduleinfo(
name="electra-small",
version="1.0.1",
summary=
"electra-small, 12-layer, 256-hidden, 4-heads, 14M parameters. The module is executed as paddle.dygraph.",
summary="electra-small, 12-layer, 256-hidden, 4-heads, 14M parameters. The module is executed as paddle.dygraph.",
author="paddlepaddle",
author_email="",
type="nlp/semantic_model",
......@@ -60,37 +59,27 @@ class Electra(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = ElectraForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='electra-small',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='electra-small', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = ElectraForTokenClassification.from_pretrained(
pretrained_model_name_or_path='electra-small',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='electra-small', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = ElectraModel.from_pretrained(pretrained_model_name_or_path='electra-small', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = ElectraModel.from_pretrained(pretrained_model_name_or_path='electra-small', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +127,7 @@ class Electra(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -181,5 +169,4 @@ class Electra(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return ElectraTokenizer.from_pretrained(
pretrained_model_name_or_path='electra-small', *args, **kwargs)
return ElectraTokenizer.from_pretrained(pretrained_model_name_or_path='electra-small', *args, **kwargs)
modules/text/language_model/ernie/module.py
浏览文件 @ d6b6c6ff
......@@ -60,29 +60,27 @@ class Ernie(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = ErnieForSequenceClassification.from_pretrained(pretrained_model_name_or_path='ernie-1.0', num_classes=self.num_classes, **kwargs)
self.model = ErnieForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='ernie-1.0', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = ErnieForTokenClassification.from_pretrained(pretrained_model_name_or_path='ernie-1.0', num_classes=self.num_classes, **kwargs)
self.model = ErnieForTokenClassification.from_pretrained(
pretrained_model_name_or_path='ernie-1.0', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = ErnieModel.from_pretrained(pretrained_model_name_or_path='ernie-1.0', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = ErnieModel.from_pretrained(pretrained_model_name_or_path='ernie-1.0', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -130,8 +128,7 @@ class Ernie(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -173,5 +170,4 @@ class Ernie(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return ErnieTokenizer.from_pretrained(
pretrained_model_name_or_path='ernie-1.0', *args, **kwargs)
return ErnieTokenizer.from_pretrained(pretrained_model_name_or_path='ernie-1.0', *args, **kwargs)
modules/text/language_model/ernie_tiny/module.py
浏览文件 @ d6b6c6ff
......@@ -59,29 +59,27 @@ class ErnieTiny(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = ErnieForSequenceClassification.from_pretrained(pretrained_model_name_or_path='ernie-tiny', num_classes=self.num_classes, **kwargs)
self.model = ErnieForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='ernie-tiny', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = ErnieForTokenClassification.from_pretrained(pretrained_model_name_or_path='ernie-tiny', num_classes=self.num_classes, **kwargs)
self.model = ErnieForTokenClassification.from_pretrained(
pretrained_model_name_or_path='ernie-tiny', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = ErnieModel.from_pretrained(pretrained_model_name_or_path='ernie-tiny', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = ErnieModel.from_pretrained(pretrained_model_name_or_path='ernie-tiny', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -129,8 +127,7 @@ class ErnieTiny(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -172,5 +169,4 @@ class ErnieTiny(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return ErnieTinyTokenizer.from_pretrained(
pretrained_model_name_or_path='ernie-tiny', *args, **kwargs)
return ErnieTinyTokenizer.from_pretrained(pretrained_model_name_or_path='ernie-tiny', *args, **kwargs)
modules/text/language_model/ernie_v2_eng_base/module.py
浏览文件 @ d6b6c6ff
......@@ -60,37 +60,27 @@ class ErnieV2(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = ErnieForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='ernie-2.0-en',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='ernie-2.0-en', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = ErnieForTokenClassification.from_pretrained(
pretrained_model_name_or_path='ernie-2.0-en',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='ernie-2.0-en', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = ErnieModel.from_pretrained(pretrained_model_name_or_path='ernie-2.0-en', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = ErnieModel.from_pretrained(pretrained_model_name_or_path='ernie-2.0-en', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +128,7 @@ class ErnieV2(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -181,5 +170,4 @@ class ErnieV2(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return ErnieTokenizer.from_pretrained(
pretrained_model_name_or_path='ernie-2.0-en', *args, **kwargs)
\ No newline at end of file
return ErnieTokenizer.from_pretrained(pretrained_model_name_or_path='ernie-2.0-en', *args, **kwargs)
modules/text/language_model/ernie_v2_eng_large/module.py
浏览文件 @ d6b6c6ff
......@@ -60,37 +60,27 @@ class ErnieV2(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = ErnieForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='ernie-2.0-large-en',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='ernie-2.0-large-en', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = ErnieForTokenClassification.from_pretrained(
pretrained_model_name_or_path='ernie-2.0-large-en',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='ernie-2.0-large-en', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = ErnieModel.from_pretrained(pretrained_model_name_or_path='ernie-2.0-large-en', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = ErnieModel.from_pretrained(pretrained_model_name_or_path='ernie-2.0-large-en', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +128,7 @@ class ErnieV2(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -181,5 +170,4 @@ class ErnieV2(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return ErnieTokenizer.from_pretrained(
pretrained_model_name_or_path='ernie-2.0-large-en', *args, **kwargs)
return ErnieTokenizer.from_pretrained(pretrained_model_name_or_path='ernie-2.0-large-en', *args, **kwargs)
modules/text/language_model/rbt3/module.py
浏览文件 @ d6b6c6ff
......@@ -60,37 +60,27 @@ class Roberta(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = RobertaForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='rbt3',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='rbt3', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = RobertaForTokenClassification.from_pretrained(
pretrained_model_name_or_path='rbt3',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='rbt3', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = RobertaModel.from_pretrained(pretrained_model_name_or_path='rbt3', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = RobertaModel.from_pretrained(pretrained_model_name_or_path='rbt3', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +128,7 @@ class Roberta(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -181,5 +170,4 @@ class Roberta(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return RobertaTokenizer.from_pretrained(
pretrained_model_name_or_path='rbt3', *args, **kwargs)
return RobertaTokenizer.from_pretrained(pretrained_model_name_or_path='rbt3', *args, **kwargs)
modules/text/language_model/rbtl3/module.py
浏览文件 @ d6b6c6ff
......@@ -60,37 +60,27 @@ class Roberta(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = RobertaForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='rbtl3',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='rbtl3', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = RobertaForTokenClassification.from_pretrained(
pretrained_model_name_or_path='rbtl3',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='rbtl3', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = RobertaModel.from_pretrained(pretrained_model_name_or_path='rbtl3', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = RobertaModel.from_pretrained(pretrained_model_name_or_path='rbtl3', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -138,8 +128,7 @@ class Roberta(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -181,5 +170,4 @@ class Roberta(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return RobertaTokenizer.from_pretrained(
pretrained_model_name_or_path='rbtl3', *args, **kwargs)
return RobertaTokenizer.from_pretrained(pretrained_model_name_or_path='rbtl3', *args, **kwargs)
modules/text/language_model/roberta-wwm-ext-large/module.py
浏览文件 @ d6b6c6ff
......@@ -61,37 +61,27 @@ class Roberta(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = RobertaForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='roberta-wwm-ext-large',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='roberta-wwm-ext-large', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = RobertaForTokenClassification.from_pretrained(
pretrained_model_name_or_path='roberta-wwm-ext-large',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='roberta-wwm-ext-large', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = RobertaModel.from_pretrained(pretrained_model_name_or_path='roberta-wwm-ext-large', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = RobertaModel.from_pretrained(pretrained_model_name_or_path='roberta-wwm-ext-large', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -139,8 +129,7 @@ class Roberta(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -182,5 +171,4 @@ class Roberta(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return RobertaTokenizer.from_pretrained(
pretrained_model_name_or_path='roberta-wwm-ext-large', *args, **kwargs)
return RobertaTokenizer.from_pretrained(pretrained_model_name_or_path='roberta-wwm-ext-large', *args, **kwargs)
modules/text/language_model/roberta-wwm-ext/module.py
浏览文件 @ d6b6c6ff
......@@ -61,37 +61,27 @@ class Roberta(nn.Layer):
task = 'seq-cls'
logger.warning(
"current task name 'sequence_classification' was renamed to 'seq-cls', "
"'sequence_classification' has been deprecated and will be removed in the future.",
)
"'sequence_classification' has been deprecated and will be removed in the future.", )
if task == 'seq-cls':
self.model = RobertaForSequenceClassification.from_pretrained(
pretrained_model_name_or_path='roberta-wwm-ext',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='roberta-wwm-ext', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task == 'token-cls':
self.model = RobertaForTokenClassification.from_pretrained(
pretrained_model_name_or_path='roberta-wwm-ext',
num_classes=self.num_classes,
**kwargs
)
pretrained_model_name_or_path='roberta-wwm-ext', num_classes=self.num_classes, **kwargs)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = ChunkEvaluator(
label_list=[self.label_map[i] for i in sorted(self.label_map.keys())]
)
self.metric = ChunkEvaluator(label_list=[self.label_map[i] for i in sorted(self.label_map.keys())])
elif task == 'text-matching':
self.model = RobertaModel.from_pretrained(pretrained_model_name_or_path='roberta-wwm-ext', **kwargs)
self.dropout = paddle.nn.Dropout(0.1)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size']*3, 2)
self.classifier = paddle.nn.Linear(self.model.config['hidden_size'] * 3, 2)
self.criterion = paddle.nn.loss.CrossEntropyLoss()
self.metric = paddle.metric.Accuracy()
elif task is None:
self.model = RobertaModel.from_pretrained(pretrained_model_name_or_path='roberta-wwm-ext', **kwargs)
else:
raise RuntimeError("Unknown task {}, task should be one in {}".format(
task, self._tasks_supported))
raise RuntimeError("Unknown task {}, task should be one in {}".format(task, self._tasks_supported))
self.task = task
......@@ -139,8 +129,7 @@ class Roberta(nn.Layer):
loss = self.criterion(logits, labels.unsqueeze(-1))
num_infer_chunks, num_label_chunks, num_correct_chunks = \
self.metric.compute(None, seq_lengths, preds, labels)
self.metric.update(
num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
self.metric.update(num_infer_chunks.numpy(), num_label_chunks.numpy(), num_correct_chunks.numpy())
_, _, f1_score = map(float, self.metric.accumulate())
return token_level_probs, loss, {'f1_score': f1_score}
return token_level_probs
......@@ -182,5 +171,4 @@ class Roberta(nn.Layer):
"""
Gets the tokenizer that is customized for this module.
"""
return RobertaTokenizer.from_pretrained(
pretrained_model_name_or_path='roberta-wwm-ext', *args, **kwargs)
return RobertaTokenizer.from_pretrained(pretrained_model_name_or_path='roberta-wwm-ext', *args, **kwargs)
modules/text/language_model/simnet_bow/module.py
浏览文件 @ d6b6c6ff
......@@ -29,7 +29,8 @@ class DataFormatError(Exception):
self.args = args
@moduleinfo(name="simnet_bow",
@moduleinfo(
name="simnet_bow",
version="1.2.0",
summary="Baidu's open-source similarity network model based on bow_pairwise.",
author="baidu-nlp",
......@@ -106,11 +107,11 @@ class SimnetBow(hub.Module):
seq_len_used = fluid.layers.squeeze(seq_len, axes=[1])
# Add embedding layer.
w_param_attrs = fluid.ParamAttr(name="emb",
initializer=fluid.initializer.TruncatedNormal(scale=0.02),
trainable=trainable)
w_param_attrs = fluid.ParamAttr(
name="emb", initializer=fluid.initializer.TruncatedNormal(scale=0.02), trainable=trainable)
dict_dim = 500002
emb_1 = fluid.layers.embedding(input=text_1,
emb_1 = fluid.layers.embedding(
input=text_1,
size=[dict_dim, 128],
is_sparse=True,
padding_idx=dict_dim - 1,
......@@ -122,7 +123,8 @@ class SimnetBow(hub.Module):
if num_slots > 1:
text_2 = fluid.data(name='text_2', shape=[-1, max_seq_len], dtype='int64', lod_level=0)
emb_2 = fluid.embedding(input=text_2,
emb_2 = fluid.embedding(
input=text_2,
size=[dict_dim, 128],
is_sparse=True,
padding_idx=dict_dim - 1,
......@@ -134,7 +136,8 @@ class SimnetBow(hub.Module):
if num_slots > 2:
text_3 = fluid.data(name='text_3', shape=[-1, max_seq_len], dtype='int64', lod_level=0)
emb_3 = fluid.embedding(input=text_3,
emb_3 = fluid.embedding(
input=text_3,
size=[dict_dim, 128],
is_sparse=True,
padding_idx=dict_dim - 1,
......@@ -295,10 +298,8 @@ class SimnetBow(hub.Module):
"""
Run as a command
"""
self.parser = argparse.ArgumentParser(description="Run the simnet_bow module.",
prog='hub run simnet_bow',
usage='%(prog)s',
add_help=True)
self.parser = argparse.ArgumentParser(
description="Run the simnet_bow module.", prog='hub run simnet_bow', usage='%(prog)s', add_help=True)
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
......@@ -323,10 +324,8 @@ class SimnetBow(hub.Module):
"""
Add the command config options
"""
self.arg_config_group.add_argument('--use_gpu',
type=ast.literal_eval,
default=False,
help="whether use GPU for prediction")
self.arg_config_group.add_argument(
'--use_gpu', type=ast.literal_eval, default=False, help="whether use GPU for prediction")
self.arg_config_group.add_argument('--batch_size', type=int, default=1, help="batch size for prediction")
......
modules/text/text_generation/Rumor_prediction/module.py
浏览文件 @ d6b6c6ff
......@@ -27,12 +27,12 @@ import numpy as np
import paddle
import paddlehub as hub
@moduleinfo(
name="Rumor_prediction",
version="1.0.0",
type="nlp/semantic_model",
summary=
"Is the input text prediction a rumor",
summary="Is the input text prediction a rumor",
author="彭兆帅，郑博培",
author_email="1084667371@qq.com，2733821739@qq.com")
class Rumorprediction(hub.Module):
......@@ -51,6 +51,7 @@ class Rumorprediction(hub.Module):
image (list(numpy.ndarray)): images data, shape of each is [H, W, C], the color space is BGR.
use_gpu(bool): Weather to use gpu
"""
# 获取数据
def get_data(sentence):
# 读取数据字典
......@@ -66,6 +67,7 @@ class Rumorprediction(hub.Module):
s = '<unk>'
data.append(int(dict_txt[s]))
return data
data = []
for text in texts:
text = get_data(text)
......@@ -75,16 +77,14 @@ class Rumorprediction(hub.Module):
place = paddle.CUDAPlace(0) if use_gpu else paddle.CPUPlace()
exe = paddle.static.Executor(place)
exe.run(paddle.static.default_startup_program())
[infer_program, feeded_var_names, target_var] = paddle.fluid.io.load_inference_model(dirname=self.default_pretrained_model_path, executor=exe)
[infer_program, feeded_var_names, target_var] = paddle.fluid.io.load_inference_model(
dirname=self.default_pretrained_model_path, executor=exe)
# 生成预测数据
tensor_words = paddle.fluid.create_lod_tensor(data, base_shape, place)
# 执行预测
result = exe.run(program=infer_program,
feed={feeded_var_names[0]: tensor_words},
fetch_list=target_var)
result = exe.run(program=infer_program, feed={feeded_var_names[0]: tensor_words}, fetch_list=target_var)
# 分类名称
names = [ '谣言', '非谣言']
names = ['谣言', '非谣言']
results = []
......@@ -93,36 +93,25 @@ class Rumorprediction(hub.Module):
content = texts[i]
lab = np.argsort(result)[0][i][-1]
alltext = {
'content': content,
'prediction': names[lab],
'probability': result[0][i][lab]
}
alltext = {'content': content, 'prediction': names[lab], 'probability': result[0][i][lab]}
alltext = [alltext]
results = results + alltext
return results
def add_module_config_arg(self):
"""
Add the command config options
"""
self.arg_config_group.add_argument(
'--use_gpu',
type=ast.literal_eval,
default=False,
help="whether use GPU for prediction")
'--use_gpu', type=ast.literal_eval, default=False, help="whether use GPU for prediction")
def add_module_input_arg(self):
"""
Add the command input options
"""
self.arg_input_group.add_argument(
'--input_text',
type=str,
default=None,
help="input_text is str")
self.arg_input_group.add_argument('--input_text', type=str, default=None, help="input_text is str")
@runnable
def run_cmd(self, argvs):
"""
......@@ -134,19 +123,15 @@ class Rumorprediction(hub.Module):
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, optional.")
title="Config options", description="Run configuration for controlling module behavior, optional.")
self.add_module_config_arg()
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
input_text = [args.input_text]
results = self.Rumor(
texts=input_text, use_gpu=args.use_gpu)
results = self.Rumor(texts=input_text, use_gpu=args.use_gpu)
return results
modules/text/text_generation/ernie_gen_leave/model/decode.py
浏览文件 @ d6b6c6ff
......@@ -23,23 +23,14 @@ import paddle.fluid.dygraph as D
def gen_bias(encoder_inputs, decoder_inputs, step):
decoder_bsz, decoder_seqlen = decoder_inputs.shape[:2]
attn_bias = L.reshape(
L.range(0, decoder_seqlen, 1, dtype='float32') + 1, [1, -1, 1])
decoder_bias = L.cast(
(L.matmul(attn_bias, 1. / attn_bias, transpose_y=True) >= 1.),
attn_bias = L.reshape(L.range(0, decoder_seqlen, 1, dtype='float32') + 1, [1, -1, 1])
decoder_bias = L.cast((L.matmul(attn_bias, 1. / attn_bias, transpose_y=True) >= 1.),
'float32') #[1, 1, decoderlen, decoderlen]
encoder_bias = L.unsqueeze(
L.cast(L.ones_like(encoder_inputs), 'float32'),
[1]) #[bsz, 1, encoderlen]
encoder_bias = L.expand(
encoder_bias, [1, decoder_seqlen, 1]) #[bsz,decoderlen, encoderlen]
decoder_bias = L.expand(decoder_bias,
[decoder_bsz, 1, 1]) #[bsz, decoderlen, decoderlen]
encoder_bias = L.unsqueeze(L.cast(L.ones_like(encoder_inputs), 'float32'), [1]) #[bsz, 1, encoderlen]
encoder_bias = L.expand(encoder_bias, [1, decoder_seqlen, 1]) #[bsz,decoderlen, encoderlen]
decoder_bias = L.expand(decoder_bias, [decoder_bsz, 1, 1]) #[bsz, decoderlen, decoderlen]
if step > 0:
bias = L.concat([
encoder_bias,
L.ones([decoder_bsz, decoder_seqlen, step], 'float32'), decoder_bias
], -1)
bias = L.concat([encoder_bias, L.ones([decoder_bsz, decoder_seqlen, step], 'float32'), decoder_bias], -1)
else:
bias = L.concat([encoder_bias, decoder_bias], -1)
return bias
......@@ -71,27 +62,16 @@ def greedy_search_infilling(model,
ids = L.stack([cls_ids, attn_ids], -1)
for step in range(max_decode_len):
bias = gen_bias(q_ids, ids, step)
pos_ids = D.to_variable(
np.tile(np.array([[step, step + 1]], dtype=np.int64), [d_batch, 1]))
pos_ids = D.to_variable(np.tile(np.array([[step, step + 1]], dtype=np.int64), [d_batch, 1]))
pos_ids += seqlen
_, logits, info = model(
ids,
L.ones_like(ids) * tgt_type_id,
pos_ids=pos_ids,
attn_bias=bias,
past_cache=past_cache)
ids, L.ones_like(ids) * tgt_type_id, pos_ids=pos_ids, attn_bias=bias, past_cache=past_cache)
gen_ids = L.argmax(logits, -1)
past_cached_k, past_cached_v = past_cache
cached_k, cached_v = info['caches']
cached_k = [
L.concat([pk, k[:, :1, :]], 1)
for pk, k in zip(past_cached_k, cached_k)
] # concat cached
cached_v = [
L.concat([pv, v[:, :1, :]], 1)
for pv, v in zip(past_cached_v, cached_v)
]
cached_k = [L.concat([pk, k[:, :1, :]], 1) for pk, k in zip(past_cached_k, cached_k)] # concat cached
cached_v = [L.concat([pv, v[:, :1, :]], 1) for pv, v in zip(past_cached_v, cached_v)]
past_cache = (cached_k, cached_v)
gen_ids = gen_ids[:, 1]
......@@ -107,10 +87,8 @@ def greedy_search_infilling(model,
return output_ids
BeamSearchState = namedtuple('BeamSearchState',
['log_probs', 'lengths', 'finished'])
BeamSearchOutput = namedtuple('BeamSearchOutput',
['scores', 'predicted_ids', 'beam_parent_ids'])
BeamSearchState = namedtuple('BeamSearchState', ['log_probs', 'lengths', 'finished'])
BeamSearchOutput = namedtuple('BeamSearchOutput', ['scores', 'predicted_ids', 'beam_parent_ids'])
def log_softmax(x):
......@@ -120,8 +98,7 @@ def log_softmax(x):
def mask_prob(p, onehot_eos, finished):
is_finished = L.cast(L.reshape(finished, [-1, 1]) != 0, 'float32')
p = is_finished * (1. - L.cast(onehot_eos, 'float32')) * -9999. + (
1. - is_finished) * p
p = is_finished * (1. - L.cast(onehot_eos, 'float32')) * -9999. + (1. - is_finished) * p
return p
......@@ -130,8 +107,7 @@ def hyp_score(log_probs, length, length_penalty):
return log_probs / lp
def beam_search_step(state, logits, eos_id, beam_width, is_first_step,
length_penalty):
def beam_search_step(state, logits, eos_id, beam_width, is_first_step, length_penalty):
"""logits.shape == [B*W, V]"""
beam_size, vocab_size = logits.shape # as batch size=1 in this hub module. the first dim means bsz * beam_size equals beam_size
logits_np = logits.numpy()
......@@ -140,8 +116,7 @@ def beam_search_step(state, logits, eos_id, beam_width, is_first_step,
logits = D.to_variable(logits_np)
bsz, beam_width = state.log_probs.shape
onehot_eos = L.cast(
F.one_hot(L.ones([1], 'int64') * eos_id, vocab_size), 'int64') #[1, V]
onehot_eos = L.cast(F.one_hot(L.ones([1], 'int64') * eos_id, vocab_size), 'int64') #[1, V]
probs = L.log(L.softmax(logits)) #[B*W, V]
probs = mask_prob(probs, onehot_eos, state.finished) #[B*W, V]
......@@ -156,32 +131,24 @@ def beam_search_step(state, logits, eos_id, beam_width, is_first_step,
alllen = L.reshape(alllen, [-1, beam_width * vocab_size])
allscore = hyp_score(allprobs, alllen, length_penalty)
if is_first_step:
allscore = L.reshape(
allscore,
[bsz, beam_width, -1])[:, 0, :] # first step only consiter beam 0
allscore = L.reshape(allscore, [bsz, beam_width, -1])[:, 0, :] # first step only consiter beam 0
scores, idx = L.topk(allscore, k=beam_width) #[B, W]
next_beam_id = idx // vocab_size #[B, W]
next_word_id = idx % vocab_size
gather_idx = L.concat([L.where(idx != -1)[:, :1],
L.reshape(idx, [-1, 1])], 1)
gather_idx = L.concat([L.where(idx != -1)[:, :1], L.reshape(idx, [-1, 1])], 1)
next_probs = L.reshape(L.gather_nd(allprobs, gather_idx), idx.shape)
next_len = L.reshape(L.gather_nd(alllen, gather_idx), idx.shape)
gather_idx = L.concat(
[L.where(next_beam_id != -1)[:, :1],
L.reshape(next_beam_id, [-1, 1])], 1)
next_finished = L.reshape(
L.gather_nd(state.finished, gather_idx),
gather_idx = L.concat([L.where(next_beam_id != -1)[:, :1], L.reshape(next_beam_id, [-1, 1])], 1)
next_finished = L.reshape(L.gather_nd(state.finished, gather_idx),
state.finished.shape) #[gather new beam state according to new beam id]
next_finished += L.cast(next_word_id == eos_id, 'int64')
next_finished = L.cast(next_finished > 0, 'int64')
next_state = BeamSearchState(
log_probs=next_probs, lengths=next_len, finished=next_finished)
output = BeamSearchOutput(
scores=scores, predicted_ids=next_word_id, beam_parent_ids=next_beam_id)
next_state = BeamSearchState(log_probs=next_probs, lengths=next_len, finished=next_finished)
output = BeamSearchOutput(scores=scores, predicted_ids=next_word_id, beam_parent_ids=next_beam_id)
return output, next_state
......@@ -210,15 +177,13 @@ def beam_search_infilling(model,
def reorder_(t, parent_id):
"""reorder cache according to parent beam id"""
gather_idx = L.where(parent_id != -1)[:, 0] * beam_width + L.reshape(
parent_id, [-1])
gather_idx = L.where(parent_id != -1)[:, 0] * beam_width + L.reshape(parent_id, [-1])
t = L.gather(t, gather_idx)
return t
def tile_(t, times):
_shapes = list(t.shape[1:])
ret = L.reshape(
L.expand(L.unsqueeze(t, [1]), [
ret = L.reshape(L.expand(L.unsqueeze(t, [1]), [
1,
times,
] + [
......@@ -241,18 +206,11 @@ def beam_search_infilling(model,
ids = L.stack([cls_ids, attn_ids], -1)
for step in range(max_decode_len):
bias = gen_bias(q_ids, ids, step)
pos_ids = D.to_variable(
np.tile(
np.array([[step, step + 1]], dtype=np.int64),
[d_batch * beam_width, 1]))
pos_ids = D.to_variable(np.tile(np.array([[step, step + 1]], dtype=np.int64), [d_batch * beam_width, 1]))
pos_ids += seqlen
_, logits, info = model(
ids,
L.ones_like(ids) * tgt_type_id,
pos_ids=pos_ids,
attn_bias=bias,
past_cache=past_cache)
ids, L.ones_like(ids) * tgt_type_id, pos_ids=pos_ids, attn_bias=bias, past_cache=past_cache)
output, state = beam_search_step(
state,
......@@ -266,17 +224,14 @@ def beam_search_infilling(model,
past_cached_k, past_cached_v = past_cache
cached_k, cached_v = info['caches']
cached_k = [
reorder_(L.concat([pk, k[:, :1, :]], 1), output.beam_parent_ids)
for pk, k in zip(past_cached_k, cached_k)
reorder_(L.concat([pk, k[:, :1, :]], 1), output.beam_parent_ids) for pk, k in zip(past_cached_k, cached_k)
] # concat cached
cached_v = [
reorder_(L.concat([pv, v[:, :1, :]], 1), output.beam_parent_ids)
for pv, v in zip(past_cached_v, cached_v)
reorder_(L.concat([pv, v[:, :1, :]], 1), output.beam_parent_ids) for pv, v in zip(past_cached_v, cached_v)
]
past_cache = (cached_k, cached_v)
pred_ids_flatten = L.reshape(output.predicted_ids,
[d_batch * beam_width])
pred_ids_flatten = L.reshape(output.predicted_ids, [d_batch * beam_width])
ids = L.stack([pred_ids_flatten, attn_ids], 1)
if state.finished.numpy().all():
......@@ -286,8 +241,7 @@ def beam_search_infilling(model,
final_parent_ids = L.stack([o.beam_parent_ids for o in outputs], 0)
final_ids = L.gather_tree(final_ids, final_parent_ids) #[:, :,
#0] #pick best beam
final_ids = L.transpose(
L.reshape(final_ids, [-1, d_batch * 1, beam_width]), [1, 2, 0])
final_ids = L.transpose(L.reshape(final_ids, [-1, d_batch * 1, beam_width]), [1, 2, 0])
return final_ids
......
modules/text/text_generation/ernie_gen_leave/model/file_utils.py
浏览文件 @ d6b6c6ff
......@@ -27,10 +27,7 @@ def _fetch_from_remote(url, force_download=False):
r = requests.get(url, stream=True)
total_len = int(r.headers.get('content-length'))
for chunk in tqdm(
r.iter_content(chunk_size=1024),
total=total_len // 1024,
desc='downloading %s' % url,
unit='KB'):
r.iter_content(chunk_size=1024), total=total_len // 1024, desc='downloading %s' % url, unit='KB'):
if chunk:
f.write(chunk)
f.flush()
......
modules/text/text_generation/ernie_gen_leave/model/modeling_ernie.py
浏览文件 @ d6b6c6ff
......@@ -30,9 +30,7 @@ def _build_linear(n_in, n_out, name, init, act=None):
return D.Linear(
n_in,
n_out,
param_attr=F.ParamAttr(
name='%s.w_0' % name if name is not None else None,
initializer=init),
param_attr=F.ParamAttr(name='%s.w_0' % name if name is not None else None, initializer=init),
bias_attr='%s.b_0' % name if name is not None else None,
act=act)
......@@ -41,11 +39,9 @@ def _build_ln(n_in, name):
return D.LayerNorm(
normalized_shape=n_in,
param_attr=F.ParamAttr(
name='%s_layer_norm_scale' % name if name is not None else None,
initializer=F.initializer.Constant(1.)),
name='%s_layer_norm_scale' % name if name is not None else None, initializer=F.initializer.Constant(1.)),
bias_attr=F.ParamAttr(
name='%s_layer_norm_bias' % name if name is not None else None,
initializer=F.initializer.Constant(1.)),
name='%s_layer_norm_bias' % name if name is not None else None, initializer=F.initializer.Constant(1.)),
)
......@@ -61,25 +57,18 @@ def append_name(name, postfix):
class AttentionLayer(D.Layer):
def __init__(self, cfg, name=None):
super(AttentionLayer, self).__init__()
initializer = F.initializer.TruncatedNormal(
scale=cfg['initializer_range'])
initializer = F.initializer.TruncatedNormal(scale=cfg['initializer_range'])
d_model = cfg['hidden_size']
n_head = cfg['num_attention_heads']
assert d_model % n_head == 0
d_model_q = cfg.get('query_hidden_size_per_head',
d_model // n_head) * n_head
d_model_v = cfg.get('value_hidden_size_per_head',
d_model // n_head) * n_head
d_model_q = cfg.get('query_hidden_size_per_head', d_model // n_head) * n_head
d_model_v = cfg.get('value_hidden_size_per_head', d_model // n_head) * n_head
self.n_head = n_head
self.d_key = d_model_q // n_head
self.q = _build_linear(d_model, d_model_q, append_name(
name, 'query_fc'), initializer)
self.k = _build_linear(d_model, d_model_q, append_name(name, 'key_fc'),
initializer)
self.v = _build_linear(d_model, d_model_v, append_name(
name, 'value_fc'), initializer)
self.o = _build_linear(d_model_v, d_model, append_name(
name, 'output_fc'), initializer)
self.q = _build_linear(d_model, d_model_q, append_name(name, 'query_fc'), initializer)
self.k = _build_linear(d_model, d_model_q, append_name(name, 'key_fc'), initializer)
self.v = _build_linear(d_model, d_model_v, append_name(name, 'value_fc'), initializer)
self.o = _build_linear(d_model_v, d_model, append_name(name, 'output_fc'), initializer)
self.dropout = lambda i: L.dropout(
i,
dropout_prob=cfg['attention_probs_dropout_prob'],
......@@ -99,14 +88,11 @@ class AttentionLayer(D.Layer):
k = L.concat([cached_k, k], 1)
v = L.concat([cached_v, v], 1)
q = L.transpose(
L.reshape(q, [0, 0, self.n_head, q.shape[-1] // self.n_head]),
q = L.transpose(L.reshape(q, [0, 0, self.n_head, q.shape[-1] // self.n_head]),
[0, 2, 1, 3]) #[batch, head, seq, dim]
k = L.transpose(
L.reshape(k, [0, 0, self.n_head, k.shape[-1] // self.n_head]),
k = L.transpose(L.reshape(k, [0, 0, self.n_head, k.shape[-1] // self.n_head]),
[0, 2, 1, 3]) #[batch, head, seq, dim]
v = L.transpose(
L.reshape(v, [0, 0, self.n_head, v.shape[-1] // self.n_head]),
v = L.transpose(L.reshape(v, [0, 0, self.n_head, v.shape[-1] // self.n_head]),
[0, 2, 1, 3]) #[batch, head, seq, dim]
q = L.scale(q, scale=self.d_key**-0.5)
......@@ -127,19 +113,12 @@ class AttentionLayer(D.Layer):
class PositionwiseFeedForwardLayer(D.Layer):
def __init__(self, cfg, name=None):
super(PositionwiseFeedForwardLayer, self).__init__()
initializer = F.initializer.TruncatedNormal(
scale=cfg['initializer_range'])
initializer = F.initializer.TruncatedNormal(scale=cfg['initializer_range'])
d_model = cfg['hidden_size']
d_ffn = cfg.get('intermediate_size', 4 * d_model)
assert cfg['hidden_act'] in ['relu', 'gelu']
self.i = _build_linear(
d_model,
d_ffn,
append_name(name, 'fc_0'),
initializer,
act=cfg['hidden_act'])
self.o = _build_linear(d_ffn, d_model, append_name(name, 'fc_1'),
initializer)
self.i = _build_linear(d_model, d_ffn, append_name(name, 'fc_0'), initializer, act=cfg['hidden_act'])
self.o = _build_linear(d_ffn, d_model, append_name(name, 'fc_1'), initializer)
prob = cfg.get('intermediate_dropout_prob', 0.)
self.dropout = lambda i: L.dropout(
i,
......@@ -158,14 +137,11 @@ class ErnieBlock(D.Layer):
def __init__(self, cfg, name=None):
super(ErnieBlock, self).__init__()
d_model = cfg['hidden_size']
initializer = F.initializer.TruncatedNormal(
scale=cfg['initializer_range'])
initializer = F.initializer.TruncatedNormal(scale=cfg['initializer_range'])
self.attn = AttentionLayer(
cfg, name=append_name(name, 'multi_head_att'))
self.attn = AttentionLayer(cfg, name=append_name(name, 'multi_head_att'))
self.ln1 = _build_ln(d_model, name=append_name(name, 'post_att'))
self.ffn = PositionwiseFeedForwardLayer(
cfg, name=append_name(name, 'ffn'))
self.ffn = PositionwiseFeedForwardLayer(cfg, name=append_name(name, 'ffn'))
self.ln2 = _build_ln(d_model, name=append_name(name, 'post_ffn'))
prob = cfg.get('intermediate_dropout_prob', cfg['hidden_dropout_prob'])
self.dropout = lambda i: L.dropout(
......@@ -175,9 +151,7 @@ class ErnieBlock(D.Layer):
) if self.training else i
def forward(self, inputs, attn_bias=None, past_cache=None):
attn_out, cache = self.attn(
inputs, inputs, inputs, attn_bias,
past_cache=past_cache) #self attn
attn_out, cache = self.attn(inputs, inputs, inputs, attn_bias, past_cache=past_cache) #self attn
attn_out = self.dropout(attn_out)
hidden = attn_out + inputs
hidden = self.ln1(hidden) # dropout/ add/ norm
......@@ -193,17 +167,13 @@ class ErnieEncoderStack(D.Layer):
def __init__(self, cfg, name=None):
super(ErnieEncoderStack, self).__init__()
n_layers = cfg['num_hidden_layers']
self.block = D.LayerList([
ErnieBlock(cfg, append_name(name, 'layer_%d' % i))
for i in range(n_layers)
])
self.block = D.LayerList([ErnieBlock(cfg, append_name(name, 'layer_%d' % i)) for i in range(n_layers)])
def forward(self, inputs, attn_bias=None, past_cache=None):
if past_cache is not None:
assert isinstance(
past_cache, tuple
), 'unknown type of `past_cache`, expect tuple or list. got %s' % repr(
type(past_cache))
past_cache,
tuple), 'unknown type of `past_cache`, expect tuple or list. got %s' % repr(type(past_cache))
past_cache = list(zip(*past_cache))
else:
past_cache = [None] * len(self.block)
......@@ -233,24 +203,18 @@ class ErnieModel(D.Layer):
d_sent = cfg.get("sent_type_vocab_size") or cfg['type_vocab_size']
self.n_head = cfg['num_attention_heads']
self.return_additional_info = cfg.get('return_additional_info', False)
initializer = F.initializer.TruncatedNormal(
scale=cfg['initializer_range'])
initializer = F.initializer.TruncatedNormal(scale=cfg['initializer_range'])
self.ln = _build_ln(d_model, name=append_name(name, 'pre_encoder'))
self.word_emb = D.Embedding([d_vocab, d_emb],
param_attr=F.ParamAttr(
name=append_name(
name, 'word_embedding'),
initializer=initializer))
name=append_name(name, 'word_embedding'), initializer=initializer))
self.pos_emb = D.Embedding([d_pos, d_emb],
param_attr=F.ParamAttr(
name=append_name(name, 'pos_embedding'),
initializer=initializer))
name=append_name(name, 'pos_embedding'), initializer=initializer))
self.sent_emb = D.Embedding([d_sent, d_emb],
param_attr=F.ParamAttr(
name=append_name(
name, 'sent_embedding'),
initializer=initializer))
name=append_name(name, 'sent_embedding'), initializer=initializer))
prob = cfg['hidden_dropout_prob']
self.dropout = lambda i: L.dropout(
i,
......@@ -258,15 +222,10 @@ class ErnieModel(D.Layer):
dropout_implementation="upscale_in_train",
) if self.training else i
self.encoder_stack = ErnieEncoderStack(cfg, append_name(
name, 'encoder'))
self.encoder_stack = ErnieEncoderStack(cfg, append_name(name, 'encoder'))
if cfg.get('has_pooler', True):
self.pooler = _build_linear(
cfg['hidden_size'],
cfg['hidden_size'],
append_name(name, 'pooled_fc'),
initializer,
act='tanh')
cfg['hidden_size'], cfg['hidden_size'], append_name(name, 'pooled_fc'), initializer, act='tanh')
else:
self.pooler = None
self.train()
......@@ -317,10 +276,7 @@ class ErnieModel(D.Layer):
encoded(`Variable` of shape `[batch_size, seq_len, hidden_size]`):
output logits of transformer stack
"""
assert len(
src_ids.shape
) == 2, 'expect src_ids.shape = [batch, sequecen], got %s' % (repr(
src_ids.shape))
assert len(src_ids.shape) == 2, 'expect src_ids.shape = [batch, sequecen], got %s' % (repr(src_ids.shape))
assert attn_bias is not None if past_cache else True, 'if `past_cache` is specified; attn_bias should not be None'
d_batch = L.shape(src_ids)[0]
d_seqlen = L.shape(src_ids)[1]
......@@ -334,21 +290,14 @@ class ErnieModel(D.Layer):
input_mask = L.unsqueeze(input_mask, axes=[-1])
attn_bias = L.matmul(input_mask, input_mask, transpose_y=True)
if use_causal_mask:
sequence = L.reshape(
L.range(0, d_seqlen, 1, dtype='float32') + 1.,
[1, 1, -1, 1])
causal_mask = L.cast(
(L.matmul(sequence, 1. / sequence, transpose_y=True) >= 1.),
'float32')
sequence = L.reshape(L.range(0, d_seqlen, 1, dtype='float32') + 1., [1, 1, -1, 1])
causal_mask = L.cast((L.matmul(sequence, 1. / sequence, transpose_y=True) >= 1.), 'float32')
attn_bias *= causal_mask
else:
assert len(
attn_bias.shape
) == 3, 'expect attn_bias tobe rank 3, got %r' % attn_bias.shape
assert len(attn_bias.shape) == 3, 'expect attn_bias tobe rank 3, got %r' % attn_bias.shape
attn_bias = (1. - attn_bias) * -10000.0
attn_bias = L.unsqueeze(attn_bias, [1])
attn_bias = L.expand(attn_bias,
[1, self.n_head, 1, 1]) # avoid broadcast =_=
attn_bias = L.expand(attn_bias, [1, self.n_head, 1, 1]) # avoid broadcast =_=
attn_bias.stop_gradient = True
if sent_ids is None:
......@@ -361,8 +310,7 @@ class ErnieModel(D.Layer):
embedded = self.dropout(self.ln(embedded))
encoded, hidden_list, cache_list = self.encoder_stack(
embedded, attn_bias, past_cache=past_cache)
encoded, hidden_list, cache_list = self.encoder_stack(embedded, attn_bias, past_cache=past_cache)
if self.pooler is not None:
pooled = self.pooler(encoded[:, 0, :])
else:
......
modules/text/text_generation/ernie_gen_leave/model/modeling_ernie_gen.py
浏览文件 @ d6b6c6ff
......@@ -24,25 +24,18 @@ class ErnieModelForGeneration(ErnieModel):
cfg['return_additional_info'] = True
cfg['has_pooler'] = False
super(ErnieModelForGeneration, self).__init__(cfg, name=name)
initializer = F.initializer.TruncatedNormal(
scale=cfg['initializer_range'])
initializer = F.initializer.TruncatedNormal(scale=cfg['initializer_range'])
d_model = cfg['hidden_size']
d_vocab = cfg['vocab_size']
self.mlm = _build_linear(
d_model,
d_model,
append_name(name, 'mask_lm_trans_fc'),
initializer,
act=cfg['hidden_act'])
self.mlm_ln = _build_ln(
d_model, name=append_name(name, 'mask_lm_trans'))
d_model, d_model, append_name(name, 'mask_lm_trans_fc'), initializer, act=cfg['hidden_act'])
self.mlm_ln = _build_ln(d_model, name=append_name(name, 'mask_lm_trans'))
self.mlm_bias = L.create_parameter(
dtype='float32',
shape=[d_vocab],
attr=F.ParamAttr(
name=append_name(name, 'mask_lm_out_fc.b_0'),
initializer=F.initializer.Constant(value=0.0)),
name=append_name(name, 'mask_lm_out_fc.b_0'), initializer=F.initializer.Constant(value=0.0)),
is_bias=True,
)
......@@ -56,23 +49,17 @@ class ErnieModelForGeneration(ErnieModel):
elif tgt_labels is None:
encoded = self.mlm(encoded)
encoded = self.mlm_ln(encoded)
logits = L.matmul(
encoded, self.word_emb.weight, transpose_y=True) + self.mlm_bias
logits = L.matmul(encoded, self.word_emb.weight, transpose_y=True) + self.mlm_bias
output_ids = L.argmax(logits, -1)
return output_ids, logits, info
else:
encoded_2d = L.gather_nd(encoded, tgt_pos)
encoded_2d = self.mlm(encoded_2d)
encoded_2d = self.mlm_ln(encoded_2d)
logits_2d = L.matmul(
encoded_2d, self.word_emb.weight,
transpose_y=True) + self.mlm_bias
logits_2d = L.matmul(encoded_2d, self.word_emb.weight, transpose_y=True) + self.mlm_bias
if len(tgt_labels.shape) == 1:
tgt_labels = L.reshape(tgt_labels, [-1, 1])
loss = L.reduce_mean(
L.softmax_with_cross_entropy(
logits_2d,
tgt_labels,
soft_label=(tgt_labels.shape[-1] != 1)))
L.softmax_with_cross_entropy(logits_2d, tgt_labels, soft_label=(tgt_labels.shape[-1] != 1)))
return loss, logits_2d, info
modules/text/text_generation/ernie_gen_leave/model/tokenizing_ernie.py
浏览文件 @ d6b6c6ff
......@@ -77,8 +77,7 @@ class ErnieTokenizer(object):
encoding='utf8',
special_token_list=[]):
if not isinstance(vocab, dict):
raise ValueError(
'expect `vocab` to be instance of dict, got %s' % type(vocab))
raise ValueError('expect `vocab` to be instance of dict, got %s' % type(vocab))
self.vocab = vocab
self.lower = lower
self.prefix = wordpiece_prefix
......@@ -89,9 +88,7 @@ class ErnieTokenizer(object):
self.unk_id = unk_token and self.vocab[unk_token]
self.mask_id = mask_token and self.vocab[mask_token]
self.unk_token = unk_token
special_tokens = {
pad_token, cls_token, sep_token, unk_token, mask_token
} | set(special_token_list)
special_tokens = {pad_token, cls_token, sep_token, unk_token, mask_token} | set(special_token_list)
pat_str = ''
for t in special_tokens:
if t is None:
......@@ -135,11 +132,9 @@ class ErnieTokenizer(object):
len2 = len(id2)
half = seqlen // 2
if len1 > len2:
len1_truncated, len2_truncated = max(half, seqlen - len2), min(
half, len2)
len1_truncated, len2_truncated = max(half, seqlen - len2), min(half, len2)
else:
len1_truncated, len2_truncated = min(half, seqlen - len1), max(
half, seqlen - len1)
len1_truncated, len2_truncated = min(half, seqlen - len1), max(half, seqlen - len1)
return id1[:len1_truncated], id2[:len2_truncated]
def build_for_ernie(self, text_id, pair_id=[]):
......@@ -155,17 +150,14 @@ class ErnieTokenizer(object):
return ret_id, ret_id_type
def encode(self, text, pair=None, truncate_to=None):
text_id = np.array(
self.convert_tokens_to_ids(self.tokenize(text)), dtype=np.int64)
text_id = np.array(self.convert_tokens_to_ids(self.tokenize(text)), dtype=np.int64)
text_id_type = np.zeros_like(text_id, dtype=np.int64)
if pair is not None:
pair_id = np.array(
self.convert_tokens_to_ids(self.tokenize(pair)), dtype=np.int64)
pair_id = np.array(self.convert_tokens_to_ids(self.tokenize(pair)), dtype=np.int64)
else:
pair_id = []
if truncate_to is not None:
text_id, pair_id = self.truncate(
text_id, [] if pair_id is None else pair_id, truncate_to)
text_id, pair_id = self.truncate(text_id, [] if pair_id is None else pair_id, truncate_to)
ret_id, ret_id_type = self.build_for_ernie(text_id, pair_id)
return ret_id, ret_id_type
modules/text/text_generation/ernie_gen_leave/module.py
浏览文件 @ d6b6c6ff
......@@ -35,8 +35,7 @@ from .model.modeling_ernie_gen import ErnieModelForGeneration
@moduleinfo(
name="ernie_gen_leave",
version="1.0.0",
summary=
"",
summary="",
author="彭兆帅，郑博培",
author_email="1084667371@qq.com，2733821739@qq.com",
type="nlp/text_generation",
......@@ -53,10 +52,7 @@ class ErnieGen(hub.NLPPredictionModule):
ernie_cfg = dict(json.loads(ernie_cfg_file.read()))
ernie_vocab_path = os.path.join(assets_path, 'vocab.txt')
with open(ernie_vocab_path, encoding='utf8') as ernie_vocab_file:
ernie_vocab = {
j.strip().split('\t')[0]: i
for i, j in enumerate(ernie_vocab_file.readlines())
}
ernie_vocab = {j.strip().split('\t')[0]: i for i, j in enumerate(ernie_vocab_file.readlines())}
with fluid.dygraph.guard(fluid.CPUPlace()):
with fluid.unique_name.guard():
......@@ -83,13 +79,10 @@ class ErnieGen(hub.NLPPredictionModule):
Returns:
results(list): the predict result.
"""
if texts and isinstance(texts, list) and all(texts) and all(
[isinstance(text, str) for text in texts]):
if texts and isinstance(texts, list) and all(texts) and all([isinstance(text, str) for text in texts]):
predicted_data = texts
else:
raise ValueError(
"The input texts should be a list with nonempty string elements."
)
raise ValueError("The input texts should be a list with nonempty string elements.")
if use_gpu and "CUDA_VISIBLE_DEVICES" not in os.environ:
use_gpu = False
......@@ -134,13 +127,9 @@ class ErnieGen(hub.NLPPredictionModule):
Add the command config options
"""
self.arg_config_group.add_argument(
'--use_gpu',
type=ast.literal_eval,
default=False,
help="whether use GPU for prediction")
'--use_gpu', type=ast.literal_eval, default=False, help="whether use GPU for prediction")
self.arg_config_group.add_argument(
'--beam_width', type=int, default=5, help="the beam search width")
self.arg_config_group.add_argument('--beam_width', type=int, default=5, help="the beam search width")
@runnable
def run_cmd(self, argvs):
......@@ -153,12 +142,9 @@ class ErnieGen(hub.NLPPredictionModule):
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, optional.")
title="Config options", description="Run configuration for controlling module behavior, optional.")
self.add_module_config_arg()
self.add_module_input_arg()
......@@ -171,7 +157,6 @@ class ErnieGen(hub.NLPPredictionModule):
self.parser.print_help()
return None
results = self.generate(
texts=input_data, use_gpu=args.use_gpu, beam_width=args.beam_width)
results = self.generate(texts=input_data, use_gpu=args.use_gpu, beam_width=args.beam_width)
return results
modules/text/text_generation/plato-mini/module.py
浏览文件 @ d6b6c6ff
......@@ -46,10 +46,8 @@ class UnifiedTransformer(nn.Layer):
"""
Convert input strings to tokens.
"""
return self.tokenizer.dialogue_encode(texts,
max_seq_len=max_seq_len,
add_start_token_as_response=True,
is_split_into_words=False)
return self.tokenizer.dialogue_encode(
texts, max_seq_len=max_seq_len, add_start_token_as_response=True, is_split_into_words=False)
def _batchify(self, data: List[List[str]], max_seq_len: int, batch_size: int):
"""
......@@ -127,7 +125,8 @@ class UnifiedTransformer(nn.Layer):
early_stopping=False,
num_return_sequences=1):
ids, scores = self.model.generate(input_ids=input_ids,
ids, scores = self.model.generate(
input_ids=input_ids,
token_type_ids=token_type_ids,
position_ids=position_ids,
attention_mask=attention_mask,
......@@ -173,11 +172,8 @@ class UnifiedTransformer(nn.Layer):
num_return_sequences = 1 if 'num_return_sequences' not in kwargs\
else kwargs['num_return_sequences']
results.extend(
select_response(ids,
scores,
self.tokenizer,
num_return_sequences=num_return_sequences,
keep_space=False))
select_response(
ids, scores, self.tokenizer, num_return_sequences=num_return_sequences, keep_space=False))
if self._interactive_mode:
self.context.append(results[0].strip())
......
modules/text/text_generation/unified_transformer-12L-cn-luge/module.py
浏览文件 @ d6b6c6ff
......@@ -46,10 +46,8 @@ class UnifiedTransformer(nn.Layer):
"""
Convert input strings to tokens.
"""
return self.tokenizer.dialogue_encode(texts,
max_seq_len=max_seq_len,
add_start_token_as_response=True,
is_split_into_words=False)
return self.tokenizer.dialogue_encode(
texts, max_seq_len=max_seq_len, add_start_token_as_response=True, is_split_into_words=False)
def _batchify(self, data: List[List[str]], max_seq_len: int, batch_size: int):
"""
......@@ -127,7 +125,8 @@ class UnifiedTransformer(nn.Layer):
early_stopping=False,
num_return_sequences=1):
ids, scores = self.model.generate(input_ids=input_ids,
ids, scores = self.model.generate(
input_ids=input_ids,
token_type_ids=token_type_ids,
position_ids=position_ids,
attention_mask=attention_mask,
......@@ -173,11 +172,8 @@ class UnifiedTransformer(nn.Layer):
num_return_sequences = 1 if 'num_return_sequences' not in kwargs\
else kwargs['num_return_sequences']
results.extend(
select_response(ids,
scores,
self.tokenizer,
num_return_sequences=num_return_sequences,
keep_space=False))
select_response(
ids, scores, self.tokenizer, num_return_sequences=num_return_sequences, keep_space=False))
if self._interactive_mode:
self.context.append(results[0].strip())
......
modules/text/text_generation/unified_transformer-12L-cn/module.py
浏览文件 @ d6b6c6ff
......@@ -46,10 +46,8 @@ class UnifiedTransformer(nn.Layer):
"""
Convert input strings to tokens.
"""
return self.tokenizer.dialogue_encode(texts,
max_seq_len=max_seq_len,
add_start_token_as_response=True,
is_split_into_words=False)
return self.tokenizer.dialogue_encode(
texts, max_seq_len=max_seq_len, add_start_token_as_response=True, is_split_into_words=False)
def _batchify(self, data: List[List[str]], max_seq_len: int, batch_size: int):
"""
......@@ -127,7 +125,8 @@ class UnifiedTransformer(nn.Layer):
early_stopping=False,
num_return_sequences=1):
ids, scores = self.model.generate(input_ids=input_ids,
ids, scores = self.model.generate(
input_ids=input_ids,
token_type_ids=token_type_ids,
position_ids=position_ids,
attention_mask=attention_mask,
......@@ -173,11 +172,8 @@ class UnifiedTransformer(nn.Layer):
num_return_sequences = 1 if 'num_return_sequences' not in kwargs\
else kwargs['num_return_sequences']
results.extend(
select_response(ids,
scores,
self.tokenizer,
num_return_sequences=num_return_sequences,
keep_space=False))
select_response(
ids, scores, self.tokenizer, num_return_sequences=num_return_sequences, keep_space=False))
if self._interactive_mode:
self.context.append(results[0].strip())
......
modules/thirdparty/image/classification/DriverStatusRecognition/module.py
浏览文件 @ d6b6c6ff
......@@ -41,18 +41,10 @@ def read_images(paths):
version='1.0.0')
class MODULE(hub.Module):
def _initialize(self, **kwargs):
self.default_pretrained_model_path = os.path.join(
self.directory, 'assets')
self.model = pdx.deploy.Predictor(self.default_pretrained_model_path,
**kwargs)
def predict(self,
images=None,
paths=None,
data=None,
batch_size=1,
use_gpu=False,
**kwargs):
self.default_pretrained_model_path = os.path.join(self.directory, 'assets')
self.model = pdx.deploy.Predictor(self.default_pretrained_model_path, **kwargs)
def predict(self, images=None, paths=None, data=None, batch_size=1, use_gpu=False, **kwargs):
all_data = images if images is not None else read_images(paths)
total_num = len(all_data)
......@@ -110,36 +102,27 @@ class MODULE(hub.Module):
prog='hub run {}'.format(self.name),
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, not required.")
title="Config options", description="Run configuration for controlling module behavior, not required.")
self.add_module_config_arg()
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
results = self.predict(
paths=[args.input_path],
use_gpu=args.use_gpu)
results = self.predict(paths=[args.input_path], use_gpu=args.use_gpu)
return results
def add_module_config_arg(self):
"""
Add the command config options.
"""
self.arg_config_group.add_argument(
'--use_gpu',
type=bool,
default=False,
help="whether use GPU or not")
self.arg_config_group.add_argument('--use_gpu', type=bool, default=False, help="whether use GPU or not")
def add_module_input_arg(self):
"""
Add the command input options.
"""
self.arg_input_group.add_argument(
'--input_path', type=str, help="path to image.")
self.arg_input_group.add_argument('--input_path', type=str, help="path to image.")
if __name__ == '__main__':
module = MODULE(directory='./new_model')
......
modules/thirdparty/image/classification/SnakeIdentification/module.py
浏览文件 @ d6b6c6ff
......@@ -41,18 +41,10 @@ def read_images(paths):
version='1.0.0')
class MODULE(hub.Module):
def _initialize(self, **kwargs):
self.default_pretrained_model_path = os.path.join(
self.directory, 'assets')
self.model = pdx.deploy.Predictor(self.default_pretrained_model_path,
**kwargs)
def predict(self,
images=None,
paths=None,
data=None,
batch_size=1,
use_gpu=False,
**kwargs):
self.default_pretrained_model_path = os.path.join(self.directory, 'assets')
self.model = pdx.deploy.Predictor(self.default_pretrained_model_path, **kwargs)
def predict(self, images=None, paths=None, data=None, batch_size=1, use_gpu=False, **kwargs):
all_data = images if images is not None else read_images(paths)
total_num = len(all_data)
......@@ -110,36 +102,27 @@ class MODULE(hub.Module):
prog='hub run {}'.format(self.name),
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, not required.")
title="Config options", description="Run configuration for controlling module behavior, not required.")
self.add_module_config_arg()
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
results = self.predict(
paths=[args.input_path],
use_gpu=args.use_gpu)
results = self.predict(paths=[args.input_path], use_gpu=args.use_gpu)
return results
def add_module_config_arg(self):
"""
Add the command config options.
"""
self.arg_config_group.add_argument(
'--use_gpu',
type=bool,
default=False,
help="whether use GPU or not")
self.arg_config_group.add_argument('--use_gpu', type=bool, default=False, help="whether use GPU or not")
def add_module_input_arg(self):
"""
Add the command input options.
"""
self.arg_input_group.add_argument(
'--input_path', type=str, help="path to image.")
self.arg_input_group.add_argument('--input_path', type=str, help="path to image.")
if __name__ == '__main__':
module = MODULE(directory='./new_model')
......
modules/thirdparty/image/classification/SpinalNet_Gemstones/gem_dataset.py
浏览文件 @ d6b6c6ff
......@@ -3,11 +3,13 @@ import numpy as np
from typing import Callable
from code.config import config_parameters
class GemStones(paddle.io.Dataset):
"""
step 1：paddle.io.Dataset
"""
def __init__(self, transforms: Callable, mode: str ='train'):
def __init__(self, transforms: Callable, mode: str = 'train'):
"""
step 2：create reader
"""
......@@ -33,7 +35,6 @@ class GemStones(paddle.io.Dataset):
elif self.mode == 'test':
self.data = test_data_folder
def __getitem__(self, index):
"""
step 3：implement __getitem__
......
modules/thirdparty/image/classification/SpinalNet_Gemstones/spinalnet_res101_gemstone/module.py
浏览文件 @ d6b6c6ff
......@@ -43,20 +43,11 @@ class BottleneckBlock(nn.Layer):
self.conv1 = nn.Conv2D(inplanes, width, 1, bias_attr=False)
self.bn1 = norm_layer(width)
self.conv2 = nn.Conv2D(width,
width,
3,
padding=dilation,
stride=stride,
groups=groups,
dilation=dilation,
bias_attr=False)
self.conv2 = nn.Conv2D(
width, width, 3, padding=dilation, stride=stride, groups=groups, dilation=dilation, bias_attr=False)
self.bn2 = norm_layer(width)
self.conv3 = nn.Conv2D(width,
planes * self.expansion,
1,
bias_attr=False)
self.conv3 = nn.Conv2D(width, planes * self.expansion, 1, bias_attr=False)
self.bn3 = norm_layer(planes * self.expansion)
self.relu = nn.ReLU()
self.downsample = downsample
......@@ -88,13 +79,7 @@ class BottleneckBlock(nn.Layer):
class ResNet(nn.Layer):
def __init__(self, block=BottleneckBlock, depth=101, with_pool=True):
super(ResNet, self).__init__()
layer_cfg = {
18: [2, 2, 2, 2],
34: [3, 4, 6, 3],
50: [3, 4, 6, 3],
101: [3, 4, 23, 3],
152: [3, 8, 36, 3]
}
layer_cfg = {18: [2, 2, 2, 2], 34: [3, 4, 6, 3], 50: [3, 4, 6, 3], 101: [3, 4, 23, 3], 152: [3, 8, 36, 3]}
layers = layer_cfg[depth]
self.with_pool = with_pool
self._norm_layer = nn.BatchNorm2D
......@@ -102,12 +87,7 @@ class ResNet(nn.Layer):
self.inplanes = 64
self.dilation = 1
self.conv1 = nn.Conv2D(3,
self.inplanes,
kernel_size=7,
stride=2,
padding=3,
bias_attr=False)
self.conv1 = nn.Conv2D(3, self.inplanes, kernel_size=7, stride=2, padding=3, bias_attr=False)
self.bn1 = self._norm_layer(self.inplanes)
self.relu = nn.ReLU()
self.maxpool = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
......@@ -127,18 +107,12 @@ class ResNet(nn.Layer):
stride = 1
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
nn.Conv2D(self.inplanes,
planes * block.expansion,
1,
stride=stride,
bias_attr=False),
nn.Conv2D(self.inplanes, planes * block.expansion, 1, stride=stride, bias_attr=False),
norm_layer(planes * block.expansion),
)
layers = []
layers.append(
block(self.inplanes, planes, stride, downsample, 1, 64,
previous_dilation, norm_layer))
layers.append(block(self.inplanes, planes, stride, downsample, 1, 64, previous_dilation, norm_layer))
self.inplanes = planes * block.expansion
for _ in range(1, blocks):
layers.append(block(self.inplanes, planes, norm_layer=norm_layer))
......@@ -161,7 +135,8 @@ class ResNet(nn.Layer):
return x
@moduleinfo(name="spinalnet_res101_gemstone",
@moduleinfo(
name="spinalnet_res101_gemstone",
type="CV/classification",
author="nanting03",
author_email="975348977@qq.com",
......@@ -194,20 +169,16 @@ class SpinalNet_ResNet101(nn.Layer):
self.half_in_size = half_in_size
self.fc_spinal_layer1 = nn.Sequential(
nn.Dropout(p=0.5), nn.Linear(half_in_size, layer_width),
nn.BatchNorm1D(layer_width), nn.ReLU())
nn.Dropout(p=0.5), nn.Linear(half_in_size, layer_width), nn.BatchNorm1D(layer_width), nn.ReLU())
self.fc_spinal_layer2 = nn.Sequential(
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width,
layer_width),
nn.BatchNorm1D(layer_width), nn.ReLU())
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width, layer_width), nn.BatchNorm1D(layer_width),
nn.ReLU())
self.fc_spinal_layer3 = nn.Sequential(
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width,
layer_width),
nn.BatchNorm1D(layer_width), nn.ReLU())
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width, layer_width), nn.BatchNorm1D(layer_width),
nn.ReLU())
self.fc_spinal_layer4 = nn.Sequential(
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width,
layer_width),
nn.BatchNorm1D(layer_width), nn.ReLU())
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width, layer_width), nn.BatchNorm1D(layer_width),
nn.ReLU())
self.fc_out = nn.Sequential(
nn.Dropout(p=0.5),
nn.Linear(layer_width * 4, class_dim),
......@@ -219,8 +190,7 @@ class SpinalNet_ResNet101(nn.Layer):
print("load custom checkpoint success")
else:
checkpoint = os.path.join(self.directory,
'spinalnet_res101.pdparams')
checkpoint = os.path.join(self.directory, 'spinalnet_res101.pdparams')
self.model_dict = paddle.load(checkpoint)
self.set_dict(self.model_dict)
print("load pretrained checkpoint success")
......@@ -240,14 +210,9 @@ class SpinalNet_ResNet101(nn.Layer):
y = paddle.flatten(y, 1)
y1 = self.fc_spinal_layer1(y[:, 0:self.half_in_size])
y2 = self.fc_spinal_layer2(
paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y1],
axis=1))
y3 = self.fc_spinal_layer3(
paddle.concat([y[:, 0:self.half_in_size], y2], axis=1))
y4 = self.fc_spinal_layer4(
paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y3],
axis=1))
y2 = self.fc_spinal_layer2(paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y1], axis=1))
y3 = self.fc_spinal_layer3(paddle.concat([y[:, 0:self.half_in_size], y2], axis=1))
y4 = self.fc_spinal_layer4(paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y3], axis=1))
y = paddle.concat([y1, y2, y3, y4], axis=1)
......
modules/thirdparty/image/classification/SpinalNet_Gemstones/spinalnet_res50_gemstone/module.py
浏览文件 @ d6b6c6ff
......@@ -43,20 +43,11 @@ class BottleneckBlock(nn.Layer):
self.conv1 = nn.Conv2D(inplanes, width, 1, bias_attr=False)
self.bn1 = norm_layer(width)
self.conv2 = nn.Conv2D(width,
width,
3,
padding=dilation,
stride=stride,
groups=groups,
dilation=dilation,
bias_attr=False)
self.conv2 = nn.Conv2D(
width, width, 3, padding=dilation, stride=stride, groups=groups, dilation=dilation, bias_attr=False)
self.bn2 = norm_layer(width)
self.conv3 = nn.Conv2D(width,
planes * self.expansion,
1,
bias_attr=False)
self.conv3 = nn.Conv2D(width, planes * self.expansion, 1, bias_attr=False)
self.bn3 = norm_layer(planes * self.expansion)
self.relu = nn.ReLU()
self.downsample = downsample
......@@ -88,13 +79,7 @@ class BottleneckBlock(nn.Layer):
class ResNet(nn.Layer):
def __init__(self, block=BottleneckBlock, depth=50, with_pool=True):
super(ResNet, self).__init__()
layer_cfg = {
18: [2, 2, 2, 2],
34: [3, 4, 6, 3],
50: [3, 4, 6, 3],
101: [3, 4, 23, 3],
152: [3, 8, 36, 3]
}
layer_cfg = {18: [2, 2, 2, 2], 34: [3, 4, 6, 3], 50: [3, 4, 6, 3], 101: [3, 4, 23, 3], 152: [3, 8, 36, 3]}
layers = layer_cfg[depth]
self.with_pool = with_pool
self._norm_layer = nn.BatchNorm2D
......@@ -102,12 +87,7 @@ class ResNet(nn.Layer):
self.inplanes = 64
self.dilation = 1
self.conv1 = nn.Conv2D(3,
self.inplanes,
kernel_size=7,
stride=2,
padding=3,
bias_attr=False)
self.conv1 = nn.Conv2D(3, self.inplanes, kernel_size=7, stride=2, padding=3, bias_attr=False)
self.bn1 = self._norm_layer(self.inplanes)
self.relu = nn.ReLU()
self.maxpool = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
......@@ -127,18 +107,12 @@ class ResNet(nn.Layer):
stride = 1
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
nn.Conv2D(self.inplanes,
planes * block.expansion,
1,
stride=stride,
bias_attr=False),
nn.Conv2D(self.inplanes, planes * block.expansion, 1, stride=stride, bias_attr=False),
norm_layer(planes * block.expansion),
)
layers = []
layers.append(
block(self.inplanes, planes, stride, downsample, 1, 64,
previous_dilation, norm_layer))
layers.append(block(self.inplanes, planes, stride, downsample, 1, 64, previous_dilation, norm_layer))
self.inplanes = planes * block.expansion
for _ in range(1, blocks):
layers.append(block(self.inplanes, planes, norm_layer=norm_layer))
......@@ -161,7 +135,8 @@ class ResNet(nn.Layer):
return x
@moduleinfo(name="spinalnet_res50_gemstone",
@moduleinfo(
name="spinalnet_res50_gemstone",
type="CV/classification",
author="nanting03",
author_email="975348977@qq.com",
......@@ -194,20 +169,16 @@ class SpinalNet_ResNet50(nn.Layer):
self.half_in_size = half_in_size
self.fc_spinal_layer1 = nn.Sequential(
nn.Dropout(p=0.5), nn.Linear(half_in_size, layer_width),
nn.BatchNorm1D(layer_width), nn.ReLU())
nn.Dropout(p=0.5), nn.Linear(half_in_size, layer_width), nn.BatchNorm1D(layer_width), nn.ReLU())
self.fc_spinal_layer2 = nn.Sequential(
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width,
layer_width),
nn.BatchNorm1D(layer_width), nn.ReLU())
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width, layer_width), nn.BatchNorm1D(layer_width),
nn.ReLU())
self.fc_spinal_layer3 = nn.Sequential(
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width,
layer_width),
nn.BatchNorm1D(layer_width), nn.ReLU())
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width, layer_width), nn.BatchNorm1D(layer_width),
nn.ReLU())
self.fc_spinal_layer4 = nn.Sequential(
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width,
layer_width),
nn.BatchNorm1D(layer_width), nn.ReLU())
nn.Dropout(p=0.5), nn.Linear(half_in_size + layer_width, layer_width), nn.BatchNorm1D(layer_width),
nn.ReLU())
self.fc_out = nn.Sequential(
nn.Dropout(p=0.5),
nn.Linear(layer_width * 4, class_dim),
......@@ -219,8 +190,7 @@ class SpinalNet_ResNet50(nn.Layer):
print("load custom checkpoint success")
else:
checkpoint = os.path.join(self.directory,
'spinalnet_res50.pdparams')
checkpoint = os.path.join(self.directory, 'spinalnet_res50.pdparams')
self.model_dict = paddle.load(checkpoint)
self.set_dict(self.model_dict)
print("load pretrained checkpoint success")
......@@ -240,14 +210,9 @@ class SpinalNet_ResNet50(nn.Layer):
y = paddle.flatten(y, 1)
y1 = self.fc_spinal_layer1(y[:, 0:self.half_in_size])
y2 = self.fc_spinal_layer2(
paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y1],
axis=1))
y3 = self.fc_spinal_layer3(
paddle.concat([y[:, 0:self.half_in_size], y2], axis=1))
y4 = self.fc_spinal_layer4(
paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y3],
axis=1))
y2 = self.fc_spinal_layer2(paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y1], axis=1))
y3 = self.fc_spinal_layer3(paddle.concat([y[:, 0:self.half_in_size], y2], axis=1))
y4 = self.fc_spinal_layer4(paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y3], axis=1))
y = paddle.concat([y1, y2, y3, y4], axis=1)
......
modules/thirdparty/image/classification/SpinalNet_Gemstones/spinalnet_vgg16_gemstone/module.py
浏览文件 @ d6b6c6ff
......@@ -61,16 +61,9 @@ def make_layers(cfg, batch_norm=False):
cfgs = {
'A': [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
'B':
[64, 64, 'M', 128, 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
'D': [
64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512, 'M', 512,
512, 512, 'M'
],
'E': [
64, 64, 'M', 128, 128, 'M', 256, 256, 256, 256, 'M', 512, 512, 512, 512,
'M', 512, 512, 512, 512, 'M'
],
'B': [64, 64, 'M', 128, 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
'D': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512, 'M', 512, 512, 512, 'M'],
'E': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 256, 'M', 512, 512, 512, 512, 'M', 512, 512, 512, 512, 'M'],
}
......@@ -86,7 +79,8 @@ def vgg16(batch_norm=False, **kwargs):
return _vgg(model_name, 'D', batch_norm, **kwargs)
@moduleinfo(name="spinalnet_vgg16_gemstone",
@moduleinfo(
name="spinalnet_vgg16_gemstone",
type="CV/classification",
author="nanting03",
author_email="975348977@qq.com",
......@@ -142,8 +136,7 @@ class SpinalNet_VGG16(nn.Layer):
nn.BatchNorm1D(layer_width),
nn.ReLU(),
)
self.fc_out = nn.Sequential(nn.Dropout(p=0.5),
nn.Linear(layer_width * 4, class_dim))
self.fc_out = nn.Sequential(nn.Dropout(p=0.5), nn.Linear(layer_width * 4, class_dim))
if load_checkpoint is not None:
self.model_dict = paddle.load(load_checkpoint)[0]
......@@ -151,8 +144,7 @@ class SpinalNet_VGG16(nn.Layer):
print("load custom checkpoint success")
else:
checkpoint = os.path.join(self.directory,
'spinalnet_vgg16.pdparams')
checkpoint = os.path.join(self.directory, 'spinalnet_vgg16.pdparams')
self.model_dict = paddle.load(checkpoint)
self.set_dict(self.model_dict)
print("load pretrained checkpoint success")
......@@ -173,14 +165,9 @@ class SpinalNet_VGG16(nn.Layer):
y = paddle.flatten(y, 1)
y1 = self.fc_spinal_layer1(y[:, 0:self.half_in_size])
y2 = self.fc_spinal_layer2(
paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y1],
axis=1))
y3 = self.fc_spinal_layer3(
paddle.concat([y[:, 0:self.half_in_size], y2], axis=1))
y4 = self.fc_spinal_layer4(
paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y3],
axis=1))
y2 = self.fc_spinal_layer2(paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y1], axis=1))
y3 = self.fc_spinal_layer3(paddle.concat([y[:, 0:self.half_in_size], y2], axis=1))
y4 = self.fc_spinal_layer4(paddle.concat([y[:, self.half_in_size:2 * self.half_in_size], y3], axis=1))
y = paddle.concat([y1, y2, y3, y4], axis=1)
......
modules/thirdparty/image/classification/food_classification/module.py
浏览文件 @ d6b6c6ff
......@@ -41,18 +41,10 @@ def read_images(paths):
version='1.0.0')
class MODULE(hub.Module):
def _initialize(self, **kwargs):
self.default_pretrained_model_path = os.path.join(
self.directory, 'assets')
self.model = pdx.deploy.Predictor(self.default_pretrained_model_path,
**kwargs)
def predict(self,
images=None,
paths=None,
data=None,
batch_size=1,
use_gpu=False,
**kwargs):
self.default_pretrained_model_path = os.path.join(self.directory, 'assets')
self.model = pdx.deploy.Predictor(self.default_pretrained_model_path, **kwargs)
def predict(self, images=None, paths=None, data=None, batch_size=1, use_gpu=False, **kwargs):
all_data = images if images is not None else read_images(paths)
total_num = len(all_data)
......@@ -110,36 +102,27 @@ class MODULE(hub.Module):
prog='hub run {}'.format(self.name),
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, not required.")
title="Config options", description="Run configuration for controlling module behavior, not required.")
self.add_module_config_arg()
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
results = self.predict(
paths=[args.input_path],
use_gpu=args.use_gpu)
results = self.predict(paths=[args.input_path], use_gpu=args.use_gpu)
return results
def add_module_config_arg(self):
"""
Add the command config options.
"""
self.arg_config_group.add_argument(
'--use_gpu',
type=bool,
default=False,
help="whether use GPU or not")
self.arg_config_group.add_argument('--use_gpu', type=bool, default=False, help="whether use GPU or not")
def add_module_input_arg(self):
"""
Add the command input options.
"""
self.arg_input_group.add_argument(
'--input_path', type=str, help="path to image.")
self.arg_input_group.add_argument('--input_path', type=str, help="path to image.")
if __name__ == '__main__':
module = MODULE(directory='./new_model')
......
modules/thirdparty/image/classification/marine_biometrics/module.py
浏览文件 @ d6b6c6ff
......@@ -37,22 +37,15 @@ def read_images(paths):
type='cv/classification',
author='郑博培、彭兆帅',
author_email='2733821739@qq.com, 1084667371@qq.com',
summary='The model uses convolution neural network to tell you the key to identify marine fish, so that anyone can call out the names of the creatures.',
summary=
'The model uses convolution neural network to tell you the key to identify marine fish, so that anyone can call out the names of the creatures.',
version='1.0.0')
class MODULE(hub.Module):
def _initialize(self, **kwargs):
self.default_pretrained_model_path = os.path.join(
self.directory, 'assets')
self.model = pdx.deploy.Predictor(self.default_pretrained_model_path,
**kwargs)
def predict(self,
images=None,
paths=None,
data=None,
batch_size=1,
use_gpu=False,
**kwargs):
self.default_pretrained_model_path = os.path.join(self.directory, 'assets')
self.model = pdx.deploy.Predictor(self.default_pretrained_model_path, **kwargs)
def predict(self, images=None, paths=None, data=None, batch_size=1, use_gpu=False, **kwargs):
all_data = images if images is not None else read_images(paths)
total_num = len(all_data)
......@@ -110,36 +103,27 @@ class MODULE(hub.Module):
prog='hub run {}'.format(self.name),
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, not required.")
title="Config options", description="Run configuration for controlling module behavior, not required.")
self.add_module_config_arg()
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
results = self.predict(
paths=[args.input_path],
use_gpu=args.use_gpu)
results = self.predict(paths=[args.input_path], use_gpu=args.use_gpu)
return results
def add_module_config_arg(self):
"""
Add the command config options.
"""
self.arg_config_group.add_argument(
'--use_gpu',
type=bool,
default=False,
help="whether use GPU or not")
self.arg_config_group.add_argument('--use_gpu', type=bool, default=False, help="whether use GPU or not")
def add_module_input_arg(self):
"""
Add the command input options.
"""
self.arg_input_group.add_argument(
'--input_path', type=str, help="path to image.")
self.arg_input_group.add_argument('--input_path', type=str, help="path to image.")
if __name__ == '__main__':
module = MODULE(directory='./new_model')
......
modules/thirdparty/image/semantic_segmentation/WatermeterSegmentation/module.py
浏览文件 @ d6b6c6ff
......@@ -34,29 +34,35 @@ def read_images(paths):
images.append(cv2.imread(path))
return images
'''旋转图像并剪裁'''
def rotate(
img, # 图片
pt1, pt2, pt3, pt4,
pt1,
pt2,
pt3,
pt4,
imgOutSrc):
# print(pt1,pt2,pt3,pt4)
withRect = math.sqrt((pt4[0] - pt1[0]) ** 2 + (pt4[1] - pt1[1]) ** 2) # 矩形框的宽度
heightRect = math.sqrt((pt1[0] - pt2[0]) ** 2 + (pt1[1] - pt2[1]) **2)
withRect = math.sqrt((pt4[0] - pt1[0])**2 + (pt4[1] - pt1[1])**2) # 矩形框的宽度
heightRect = math.sqrt((pt1[0] - pt2[0])**2 + (pt1[1] - pt2[1])**2)
# print("矩形的宽度",withRect, "矩形的高度", heightRect)
angle = acos((pt4[0] - pt1[0]) / withRect) * (180 / math.pi) # 矩形框旋转角度
# print("矩形框旋转角度", angle)
if withRect > heightRect:
if pt4[1]>pt1[1]:
if pt4[1] > pt1[1]:
pass
# print("顺时针旋转")
else:
# print("逆时针旋转")
angle=-angle
angle = -angle
else:
# print("逆时针旋转")
angle=90 - angle
angle = 90 - angle
height = img.shape[0] # 原始图像高度
width = img.shape[1] # 原始图像宽度
......@@ -76,14 +82,15 @@ def rotate(
[[pt4[0]], [pt4[1]]] = np.dot(rotateMat, np.array([[pt4[0]], [pt4[1]], [1]]))
# 处理反转的情况
if pt2[1]>pt4[1]:
pt2[1],pt4[1]=pt4[1],pt2[1]
if pt1[0]>pt3[0]:
pt1[0],pt3[0]=pt3[0],pt1[0]
if pt2[1] > pt4[1]:
pt2[1], pt4[1] = pt4[1], pt2[1]
if pt1[0] > pt3[0]:
pt1[0], pt3[0] = pt3[0], pt1[0]
imgOut = imgRotation[int(pt2[1]):int(pt4[1]), int(pt1[0]):int(pt3[0])]
cv2.imwrite(imgOutSrc, imgOut) # 裁减得到的旋转矩形框
@moduleinfo(
name='WatermeterSegmentation',
type='CV/semantic_segmentatio',
......@@ -93,18 +100,10 @@ def rotate(
version='1.0.0')
class MODULE(hub.Module):
def _initialize(self, **kwargs):
self.default_pretrained_model_path = os.path.join(
self.directory, 'assets')
self.model = pdx.deploy.Predictor(self.default_pretrained_model_path,
**kwargs)
def predict(self,
images=None,
paths=None,
data=None,
batch_size=1,
use_gpu=False,
**kwargs):
self.default_pretrained_model_path = os.path.join(self.directory, 'assets')
self.model = pdx.deploy.Predictor(self.default_pretrained_model_path, **kwargs)
def predict(self, images=None, paths=None, data=None, batch_size=1, use_gpu=False, **kwargs):
all_data = images if images is not None else read_images(paths)
total_num = len(all_data)
......@@ -134,7 +133,7 @@ class MODULE(hub.Module):
print(type(contours[0]))
n = 0
m = 0
for index,contour in enumerate(contours):
for index, contour in enumerate(contours):
if len(contour) > n:
n = len(contour)
m = index
......@@ -181,7 +180,6 @@ class MODULE(hub.Module):
res.append(result)
return res
@runnable
def run_cmd(self, argvs):
"""
......@@ -192,36 +190,27 @@ class MODULE(hub.Module):
prog='hub run {}'.format(self.name),
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, not required.")
title="Config options", description="Run configuration for controlling module behavior, not required.")
self.add_module_config_arg()
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
results = self.predict(
paths=[args.input_path],
use_gpu=args.use_gpu)
results = self.predict(paths=[args.input_path], use_gpu=args.use_gpu)
return results
def add_module_config_arg(self):
"""
Add the command config options.
"""
self.arg_config_group.add_argument(
'--use_gpu',
type=bool,
default=False,
help="whether use GPU or not")
self.arg_config_group.add_argument('--use_gpu', type=bool, default=False, help="whether use GPU or not")
def add_module_input_arg(self):
"""
Add the command input options.
"""
self.arg_input_group.add_argument(
'--input_path', type=str, help="path to image.")
self.arg_input_group.add_argument('--input_path', type=str, help="path to image.")
if __name__ == '__main__':
module = MODULE(directory='./new_model')
......
modules/thirdparty/image/text_recognition/Vehicle_License_Plate_Recognition/module.py
浏览文件 @ d6b6c6ff
......@@ -13,16 +13,13 @@ from paddlehub.module.module import moduleinfo, serving
author="jm12138",
author_email="",
summary="Vehicle_License_Plate_Recognition",
version="1.0.0"
)
version="1.0.0")
class Vehicle_License_Plate_Recognition(nn.Layer):
def __init__(self):
super(Vehicle_License_Plate_Recognition, self).__init__()
self.vlpr = PaddleOCR(
det_model_dir=os.path.join(self.directory, 'det_vlpr'),
rec_model_dir=os.path.join(
self.directory, 'ch_ppocr_server_v2.0_rec_infer')
)
rec_model_dir=os.path.join(self.directory, 'ch_ppocr_server_v2.0_rec_infer'))
@staticmethod
def base64_to_cv2(b64str):
......@@ -38,17 +35,11 @@ class Vehicle_License_Plate_Recognition(nn.Layer):
if isinstance(images, list):
for item in images:
for bbox, text in self.vlpr.ocr(item):
results.append({
'license': text[0],
'bbox': bbox
})
results.append({'license': text[0], 'bbox': bbox})
elif isinstance(images, (str, np.ndarray)):
for bbox, text in self.vlpr.ocr(images):
results.append({
'license': text[0],
'bbox': bbox
})
results.append({'license': text[0], 'bbox': bbox})
return results
......
modules/thirdparty/text/text_generation/GPT2_Base_CN/module.py
浏览文件 @ d6b6c6ff
......@@ -38,18 +38,15 @@ class GPT2_Base_CN(nn.Layer):
# 初始预测
ids = self.tokenizer.encode(text)
input_id = paddle.to_tensor(
np.array(ids).reshape(1, -1).astype('int64'))
input_id = paddle.to_tensor(np.array(ids).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True)
next_token = int(np.argmax(output[0, -1].numpy()))
ids.append(next_token)
# 使用缓存进行继续预测
for i in range(max_len-1):
input_id = paddle.to_tensor(
np.array([next_token]).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(
input_id, use_cache=True, cache=cached_kvs)
for i in range(max_len - 1):
input_id = paddle.to_tensor(np.array([next_token]).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True, cache=cached_kvs)
next_token = int(np.argmax(output[0, -1].numpy()))
ids.append(next_token)
......@@ -79,14 +76,12 @@ class GPT2_Base_CN(nn.Layer):
if top_p < 1.0:
sorted_logits = paddle.sort(logits, descending=True)
sorted_indices = paddle.argsort(logits, descending=True).numpy()
cumulative_probs = paddle.cumsum(paddle.nn.functional.softmax(
sorted_logits, axis=-1), axis=-1).numpy()
cumulative_probs = paddle.cumsum(paddle.nn.functional.softmax(sorted_logits, axis=-1), axis=-1).numpy()
# Remove tokens with cumulative probability above the threshold
sorted_indices_to_remove = cumulative_probs > top_p
# Shift the indices to the right to keep also the first token above the threshold
sorted_indices_to_remove[...,
1:] = sorted_indices_to_remove[..., :-1]
sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[..., :-1]
sorted_indices_to_remove[..., 0] = 0
indices_to_remove = sorted_indices[sorted_indices_to_remove]
......@@ -94,7 +89,14 @@ class GPT2_Base_CN(nn.Layer):
return paddle.to_tensor(logits_np)
def nucleus_sample(self, text, max_len=32, end_word=None, repitition_penalty=1.0, temperature=1.0, top_k=0, top_p=1.0):
def nucleus_sample(self,
text,
max_len=32,
end_word=None,
repitition_penalty=1.0,
temperature=1.0,
top_k=0,
top_p=1.0):
with paddle.no_grad():
# 终止标志
if end_word is not None:
......@@ -103,33 +105,28 @@ class GPT2_Base_CN(nn.Layer):
# 初始预测
ids = self.tokenizer.encode(text)
input_id = paddle.to_tensor(
np.array(ids).reshape(1, -1).astype('int64'))
input_id = paddle.to_tensor(np.array(ids).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True)
next_token_logits = output[0, -1, :]
for id in set(ids):
next_token_logits[id] /= repitition_penalty
next_token_logits = next_token_logits / temperature
filtered_logits = self.top_k_top_p_filtering(
next_token_logits, top_k=top_k, top_p=top_p)
next_token = paddle.multinomial(paddle.nn.functional.softmax(
filtered_logits, axis=-1), num_samples=1).numpy()
filtered_logits = self.top_k_top_p_filtering(next_token_logits, top_k=top_k, top_p=top_p)
next_token = paddle.multinomial(
paddle.nn.functional.softmax(filtered_logits, axis=-1), num_samples=1).numpy()
ids += [int(next_token)]
# 使用缓存进行继续预测
for i in range(max_len-1):
input_id = paddle.to_tensor(
np.array([next_token]).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(
input_id, use_cache=True, cache=cached_kvs)
for i in range(max_len - 1):
input_id = paddle.to_tensor(np.array([next_token]).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True, cache=cached_kvs)
next_token_logits = output[0, -1, :]
for id in set(ids):
next_token_logits[id] /= repitition_penalty
next_token_logits = next_token_logits / temperature
filtered_logits = self.top_k_top_p_filtering(
next_token_logits, top_k=top_k, top_p=top_p)
next_token = paddle.multinomial(paddle.nn.functional.softmax(
filtered_logits, axis=-1), num_samples=1).numpy()
filtered_logits = self.top_k_top_p_filtering(next_token_logits, top_k=top_k, top_p=top_p)
next_token = paddle.multinomial(
paddle.nn.functional.softmax(filtered_logits, axis=-1), num_samples=1).numpy()
ids += [int(next_token)]
# 根据终止标志停止预测
......
modules/thirdparty/text/text_generation/GPT2_CPM_LM/module.py
浏览文件 @ d6b6c6ff
......@@ -30,13 +30,11 @@ class GPT2_CPM_LM(nn.Layer):
max_position_embeddings=1024,
type_vocab_size=1,
initializer_range=0.02,
pad_token_id=0
)
pad_token_id=0)
self.model = GPT2ForPretraining(gpt2)
# 读取CPM-LM模型参数(FP16)
state_dict = paddle.load(os.path.join(
self.directory, 'CPM-LM.pdparams'))
state_dict = paddle.load(os.path.join(self.directory, 'CPM-LM.pdparams'))
# FP16 -> FP32
for param in state_dict:
......@@ -51,8 +49,7 @@ class GPT2_CPM_LM(nn.Layer):
# 加载编解码器
self.tokenizer = GPT2ChineseTokenizer(
vocab_file=os.path.join(self.directory, 'vocab.json'),
model_file=os.path.join(self.directory, 'chinese_vocab.model')
)
model_file=os.path.join(self.directory, 'chinese_vocab.model'))
# 初始化编码器
_ = self.tokenizer.encode('_')
......@@ -67,18 +64,15 @@ class GPT2_CPM_LM(nn.Layer):
# 初始预测
ids = self.tokenizer.encode(text)
input_id = paddle.to_tensor(
np.array(ids).reshape(1, -1).astype('int64'))
input_id = paddle.to_tensor(np.array(ids).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True)
next_token = int(np.argmax(output[0, -1].numpy()))
ids.append(next_token)
# 使用缓存进行继续预测
for i in range(max_len-1):
input_id = paddle.to_tensor(
np.array([next_token]).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(
input_id, use_cache=True, cache=cached_kvs)
for i in range(max_len - 1):
input_id = paddle.to_tensor(np.array([next_token]).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True, cache=cached_kvs)
next_token = int(np.argmax(output[0, -1].numpy()))
ids.append(next_token)
......@@ -108,14 +102,12 @@ class GPT2_CPM_LM(nn.Layer):
if top_p < 1.0:
sorted_logits = paddle.sort(logits, descending=True)
sorted_indices = paddle.argsort(logits, descending=True).numpy()
cumulative_probs = paddle.cumsum(paddle.nn.functional.softmax(
sorted_logits, axis=-1), axis=-1).numpy()
cumulative_probs = paddle.cumsum(paddle.nn.functional.softmax(sorted_logits, axis=-1), axis=-1).numpy()
# Remove tokens with cumulative probability above the threshold
sorted_indices_to_remove = cumulative_probs > top_p
# Shift the indices to the right to keep also the first token above the threshold
sorted_indices_to_remove[...,
1:] = sorted_indices_to_remove[..., :-1]
sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[..., :-1]
sorted_indices_to_remove[..., 0] = 0
indices_to_remove = sorted_indices[sorted_indices_to_remove]
......@@ -123,7 +115,14 @@ class GPT2_CPM_LM(nn.Layer):
return paddle.to_tensor(logits_np)
def nucleus_sample(self, text, max_len=32, end_word=None, repitition_penalty=1.0, temperature=1.0, top_k=0, top_p=1.0):
def nucleus_sample(self,
text,
max_len=32,
end_word=None,
repitition_penalty=1.0,
temperature=1.0,
top_k=0,
top_p=1.0):
with paddle.no_grad():
# 终止标志
if end_word is not None:
......@@ -132,35 +131,30 @@ class GPT2_CPM_LM(nn.Layer):
# 初始预测
ids = self.tokenizer.encode(text)
input_id = paddle.to_tensor(
np.array(ids).reshape(1, -1).astype('int64'))
input_id = paddle.to_tensor(np.array(ids).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True)
next_token_logits = output[0, -1, :]
for id in set(ids):
next_token_logits[id] /= repitition_penalty
next_token_logits = next_token_logits / temperature
next_token_logits[self.tokenizer.encoder['<unk>']] = -float('Inf')
filtered_logits = self.top_k_top_p_filtering(
next_token_logits, top_k=top_k, top_p=top_p)
next_token = paddle.multinomial(paddle.nn.functional.softmax(
filtered_logits, axis=-1), num_samples=1).numpy()
filtered_logits = self.top_k_top_p_filtering(next_token_logits, top_k=top_k, top_p=top_p)
next_token = paddle.multinomial(
paddle.nn.functional.softmax(filtered_logits, axis=-1), num_samples=1).numpy()
ids += [int(next_token)]
# 使用缓存进行继续预测
for i in range(max_len-1):
input_id = paddle.to_tensor(
np.array([next_token]).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(
input_id, use_cache=True, cache=cached_kvs)
for i in range(max_len - 1):
input_id = paddle.to_tensor(np.array([next_token]).reshape(1, -1).astype('int64'))
output, cached_kvs = self.model(input_id, use_cache=True, cache=cached_kvs)
next_token_logits = output[0, -1, :]
for id in set(ids):
next_token_logits[id] /= repitition_penalty
next_token_logits = next_token_logits / temperature
next_token_logits[self.tokenizer.encoder['<unk>']] = -float('Inf')
filtered_logits = self.top_k_top_p_filtering(
next_token_logits, top_k=top_k, top_p=top_p)
next_token = paddle.multinomial(paddle.nn.functional.softmax(
filtered_logits, axis=-1), num_samples=1).numpy()
filtered_logits = self.top_k_top_p_filtering(next_token_logits, top_k=top_k, top_p=top_p)
next_token = paddle.multinomial(
paddle.nn.functional.softmax(filtered_logits, axis=-1), num_samples=1).numpy()
ids += [int(next_token)]
# 根据终止标志停止预测
......
modules/thirdparty/video/Video_editing/SkyAR/module.py
浏览文件 @ d6b6c6ff
......@@ -4,41 +4,42 @@ from .skyfilter import SkyFilter
from paddlehub.module.module import moduleinfo
@moduleinfo(
name="SkyAR",
type="CV/Video_editing",
author="jm12138",
author_email="",
summary="SkyAR",
version="1.0.0"
)
@moduleinfo(name="SkyAR", type="CV/Video_editing", author="jm12138", author_email="", summary="SkyAR", version="1.0.0")
class SkyAR(nn.Layer):
def __init__(self, model_path=None):
super(SkyAR, self).__init__()
self.imgs = ['cloudy', 'district9ship', 'floatingcastle',
'galaxy', 'jupiter', 'rainy', 'sunny', 'sunset', 'supermoon']
self.imgs = [
'cloudy', 'district9ship', 'floatingcastle', 'galaxy', 'jupiter', 'rainy', 'sunny', 'sunset', 'supermoon'
]
self.videos = ['thunderstorm']
if model_path:
self.model_path = model_path
else:
self.model_path = os.path.join(self.directory, './ResNet50FCN')
def MagicSky(
self, video_path, save_path, config='jupiter',
is_rainy=False, preview_frames_num=0, is_video_sky=False, is_show=False,
skybox_img=None, skybox_video=None, rain_cap_path=None,
halo_effect=True, auto_light_matching=False,
relighting_factor=0.8, recoloring_factor=0.5, skybox_center_crop=0.5
):
def MagicSky(self,
video_path,
save_path,
config='jupiter',
is_rainy=False,
preview_frames_num=0,
is_video_sky=False,
is_show=False,
skybox_img=None,
skybox_video=None,
rain_cap_path=None,
halo_effect=True,
auto_light_matching=False,
relighting_factor=0.8,
recoloring_factor=0.5,
skybox_center_crop=0.5):
if config in self.imgs:
skybox_img = os.path.join(
self.directory, 'skybox', '%s.jpg' % config)
skybox_img = os.path.join(self.directory, 'skybox', '%s.jpg' % config)
skybox_video = None
is_video_sky = False
elif config in self.videos:
skybox_img = None
skybox_video = os.path.join(
self.directory, 'skybox', '%s.mp4' % config)
skybox_video = os.path.join(self.directory, 'skybox', '%s.mp4' % config)
is_video_sky = True
elif skybox_img:
is_video_sky = False
......@@ -49,8 +50,7 @@ class SkyAR(nn.Layer):
raise 'please check your configs'
if not rain_cap_path:
rain_cap_path = os.path.join(
self.directory, 'rain_streaks', 'videoplayback.mp4')
rain_cap_path = os.path.join(self.directory, 'rain_streaks', 'videoplayback.mp4')
skyfilter = SkyFilter(
model_path=self.model_path,
......@@ -67,7 +67,6 @@ class SkyAR(nn.Layer):
skybox_video=skybox_video,
is_video=is_video_sky,
is_rainy=is_rainy,
is_show=is_show
)
is_show=is_show)
skyfilter.run(preview_frames_num)
modules/thirdparty/video/Video_editing/SkyAR/rain.py
浏览文件 @ d6b6c6ff
......@@ -29,7 +29,7 @@ class Rain():
return rain_layer
def _create_haze_layer(self, rain_layer):
return 0.1*np.ones_like(rain_layer)
return 0.1 * np.ones_like(rain_layer)
def forward(self, img):
# get input image size
......@@ -54,7 +54,7 @@ class Rain():
img_out = 1 - (1 - rain_layer) * (1 - img)
# gamma and light correction
img_out = self.light_correction*(img_out**self.gamma)
img_out = self.light_correction * (img_out**self.gamma)
# check boundary
img_out = np.clip(img_out, a_min=0, a_max=1.)
......
modules/thirdparty/video/Video_editing/SkyAR/skybox.py
浏览文件 @ d6b6c6ff
......@@ -6,10 +6,8 @@ from .utils import build_transformation_matrix, update_transformation_matrix, es
class SkyBox():
def __init__(
self, out_size, skybox_img, skybox_video, halo_effect,
auto_light_matching, relighting_factor, recoloring_factor,
skybox_center_crop, rain_cap_path, is_video, is_rainy):
def __init__(self, out_size, skybox_img, skybox_video, halo_effect, auto_light_matching, relighting_factor,
recoloring_factor, skybox_center_crop, rain_cap_path, is_video, is_rainy):
self.out_size_w, self.out_size_h = out_size
......@@ -28,12 +26,7 @@ class SkyBox():
self.skybox_center_crop = skybox_center_crop
self.load_skybox()
self.rainmodel = Rain(
rain_cap_path=rain_cap_path,
rain_intensity=0.8,
haze_intensity=0.0,
gamma=1.0,
light_correction=1.0
)
rain_cap_path=rain_cap_path, rain_intensity=0.8, haze_intensity=0.0, gamma=1.0, light_correction=1.0)
# motion parameters
self.M = np.array([[1, 0, 0], [0, 1, 0]], dtype=np.float32)
......@@ -43,10 +36,8 @@ class SkyBox():
def tile_skybox_img(self, imgtile):
screen_y1 = int(imgtile.shape[0] / 2 - self.out_size_h / 2)
screen_x1 = int(imgtile.shape[1] / 2 - self.out_size_w / 2)
imgtile = np.concatenate(
[imgtile[screen_y1:, :, :], imgtile[0:screen_y1, :, :]], axis=0)
imgtile = np.concatenate(
[imgtile[:, screen_x1:, :], imgtile[:, 0:screen_x1, :]], axis=1)
imgtile = np.concatenate([imgtile[screen_y1:, :, :], imgtile[0:screen_y1, :, :]], axis=0)
imgtile = np.concatenate([imgtile[:, screen_x1:, :], imgtile[:, 0:screen_x1, :]], axis=1)
return imgtile
......@@ -57,12 +48,9 @@ class SkyBox():
skybox_img = cv2.imread(self.skybox_img, cv2.IMREAD_COLOR)
skybox_img = cv2.cvtColor(skybox_img, cv2.COLOR_BGR2RGB)
self.skybox_img = cv2.resize(
skybox_img, (self.out_size_w, self.out_size_h))
self.skybox_img = cv2.resize(skybox_img, (self.out_size_w, self.out_size_h))
cc = 1. / self.skybox_center_crop
imgtile = cv2.resize(
skybox_img, (int(cc * self.out_size_w),
int(cc*self.out_size_h)))
imgtile = cv2.resize(skybox_img, (int(cc * self.out_size_w), int(cc * self.out_size_h)))
self.skybox_imgx2 = self.tile_skybox_img(imgtile)
self.skybox_imgx2 = np.expand_dims(self.skybox_imgx2, axis=0)
......@@ -71,15 +59,11 @@ class SkyBox():
cap = cv2.VideoCapture(self.skybox_video)
m_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
cc = 1. / self.skybox_center_crop
self.skybox_imgx2 = np.zeros(
[m_frames, int(cc*self.out_size_h),
int(cc*self.out_size_w), 3], np.uint8)
self.skybox_imgx2 = np.zeros([m_frames, int(cc * self.out_size_h), int(cc * self.out_size_w), 3], np.uint8)
for i in range(m_frames):
_, skybox_img = cap.read()
skybox_img = cv2.cvtColor(skybox_img, cv2.COLOR_BGR2RGB)
imgtile = cv2.resize(
skybox_img, (int(cc * self.out_size_w),
int(cc * self.out_size_h)))
imgtile = cv2.resize(skybox_img, (int(cc * self.out_size_w), int(cc * self.out_size_h)))
skybox_imgx2 = self.tile_skybox_img(imgtile)
self.skybox_imgx2[i, :] = skybox_imgx2
......@@ -87,8 +71,7 @@ class SkyBox():
r, eps = 20, 0.01
refined_skymask = guidedfilter(img[:, :, 2], G_pred[:, :, 0], r, eps)
refined_skymask = np.stack(
[refined_skymask, refined_skymask, refined_skymask], axis=-1)
refined_skymask = np.stack([refined_skymask, refined_skymask, refined_skymask], axis=-1)
return np.clip(refined_skymask, a_min=0, a_max=1)
......@@ -98,14 +81,13 @@ class SkyBox():
nbgs, bgh, bgw, c = self.skybox_imgx2.shape
fetch_id = self.frame_id % nbgs
skybg_warp = cv2.warpAffine(
self.skybox_imgx2[fetch_id, :, :, :], self.M,
(bgw, bgh), borderMode=cv2.BORDER_WRAP)
self.skybox_imgx2[fetch_id, :, :, :], self.M, (bgw, bgh), borderMode=cv2.BORDER_WRAP)
skybg = skybg_warp[0:self.out_size_h, 0:self.out_size_w, :]
self.frame_id += 1
return np.array(skybg, np.float32)/255.
return np.array(skybg, np.float32) / 255.
def skybox_tracking(self, frame, frame_prev, skymask):
if np.mean(skymask) < 0.05:
......@@ -113,27 +95,25 @@ class SkyBox():
return np.array([[1, 0, 0], [0, 1, 0]], dtype=np.float32)
prev_gray = cv2.cvtColor(frame_prev, cv2.COLOR_RGB2GRAY)
prev_gray = np.array(255*prev_gray, dtype=np.uint8)
prev_gray = np.array(255 * prev_gray, dtype=np.uint8)
curr_gray = cv2.cvtColor(frame, cv2.COLOR_RGB2GRAY)
curr_gray = np.array(255*curr_gray, dtype=np.uint8)
curr_gray = np.array(255 * curr_gray, dtype=np.uint8)
mask = np.array(skymask[:, :, 0] > 0.99, dtype=np.uint8)
template_size = int(0.05*mask.shape[0])
template_size = int(0.05 * mask.shape[0])
mask = cv2.erode(mask, np.ones([template_size, template_size]))
# ShiTomasi corner detection
prev_pts = cv2.goodFeaturesToTrack(
prev_gray, mask=mask, maxCorners=200,
qualityLevel=0.01, minDistance=30, blockSize=3)
prev_gray, mask=mask, maxCorners=200, qualityLevel=0.01, minDistance=30, blockSize=3)
if prev_pts is None:
print('no feature point detected')
return np.array([[1, 0, 0], [0, 1, 0]], dtype=np.float32)
# Calculate optical flow (i.e. track feature points)
curr_pts, status, err = cv2.calcOpticalFlowPyrLK(
prev_gray, curr_gray, prev_pts, None)
curr_pts, status, err = cv2.calcOpticalFlowPyrLK(prev_gray, curr_gray, prev_pts, None)
# Filter only valid points
idx = np.where(status == 1)[0]
if idx.size == 0:
......@@ -147,15 +127,14 @@ class SkyBox():
return np.array([[1, 0, 0], [0, 1, 0]], dtype=np.float32)
# limit the motion to translation + rotation
dxdyda = estimate_partial_transform((
np.array(prev_pts), np.array(curr_pts)))
dxdyda = estimate_partial_transform((np.array(prev_pts), np.array(curr_pts)))
m = build_transformation_matrix(dxdyda)
return m
def relighting(self, img, skybg, skymask):
# color matching, reference: skybox_img
step = int(img.shape[0]/20)
step = int(img.shape[0] / 20)
skybg_thumb = skybg[::step, ::step, :]
img_thumb = img[::step, ::step, :]
skymask_thumb = skymask[::step, ::step, :]
......@@ -163,7 +142,7 @@ class SkyBox():
img_mean = np.sum(img_thumb * (1-skymask_thumb), axis=(0, 1), keepdims=True) \
/ ((1-skymask_thumb).sum(axis=(0, 1), keepdims=True) + 1e-9)
diff = skybg_mean - img_mean
img_colortune = img + self.recoloring_factor*diff
img_colortune = img + self.recoloring_factor * diff
if self.auto_light_matching:
img = img_colortune
......@@ -176,11 +155,9 @@ class SkyBox():
def halo(self, syneth, skybg, skymask):
# reflection
halo = 0.5*cv2.blur(
skybg*skymask, (int(self.out_size_w/5),
int(self.out_size_w/5)))
halo = 0.5 * cv2.blur(skybg * skymask, (int(self.out_size_w / 5), int(self.out_size_w / 5)))
# screen blend 1 - (1-a)(1-b)
syneth_with_halo = 1 - (1-syneth) * (1-halo)
syneth_with_halo = 1 - (1 - syneth) * (1 - halo)
return syneth_with_halo
......
modules/thirdparty/video/Video_editing/SkyAR/skyfilter.py
浏览文件 @ d6b6c6ff
......@@ -8,37 +8,37 @@ __all__ = ['SkyFilter']
class SkyFilter():
def __init__(self, model_path, video_path, save_path, in_size,
halo_effect, auto_light_matching, relighting_factor,
recoloring_factor, skybox_center_crop, rain_cap_path,
skybox_img, skybox_video, is_video, is_rainy, is_show
):
def __init__(self, model_path, video_path, save_path, in_size, halo_effect, auto_light_matching, relighting_factor,
recoloring_factor, skybox_center_crop, rain_cap_path, skybox_img, skybox_video, is_video, is_rainy,
is_show):
self.in_size = in_size
self.is_show = is_show
self.cap = cv2.VideoCapture(video_path)
self.m_frames = int(self.cap.get(cv2.CAP_PROP_FRAME_COUNT))
self.fps = self.cap.get(cv2.CAP_PROP_FPS)
self.out_size = int(self.cap.get(
cv2.CAP_PROP_FRAME_WIDTH)), int(self.cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
self.out_size = int(self.cap.get(cv2.CAP_PROP_FRAME_WIDTH)), int(self.cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
self.model = paddle.jit.load(
model_path, model_filename='__model__', params_filename='__params__')
self.model = paddle.jit.load(model_path, model_filename='__model__', params_filename='__params__')
self.model.eval()
self.skyboxengine = SkyBox(
out_size=self.out_size, skybox_img=skybox_img, skybox_video=skybox_video,
halo_effect=halo_effect, auto_light_matching=auto_light_matching,
relighting_factor=relighting_factor, recoloring_factor=recoloring_factor,
skybox_center_crop=skybox_center_crop, rain_cap_path=rain_cap_path,
is_video=is_video, is_rainy=is_rainy
)
out_size=self.out_size,
skybox_img=skybox_img,
skybox_video=skybox_video,
halo_effect=halo_effect,
auto_light_matching=auto_light_matching,
relighting_factor=relighting_factor,
recoloring_factor=recoloring_factor,
skybox_center_crop=skybox_center_crop,
rain_cap_path=rain_cap_path,
is_video=is_video,
is_rainy=is_rainy)
path, _ = os.path.split(save_path)
if path == '':
path = '.'
if not os.path.exists(path):
os.mkdir(path)
self.video_writer = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'MP4V'),
self.fps, self.out_size)
self.video_writer = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'MP4V'), self.fps, self.out_size)
def synthesize(self, img_HD, img_HD_prev):
h, w, _ = img_HD.shape
......@@ -50,8 +50,7 @@ class SkyFilter():
img = paddle.to_tensor(img)
G_pred = self.model(img)
G_pred = paddle.nn.functional.interpolate(
G_pred, (h, w), mode='bicubic', align_corners=False)
G_pred = paddle.nn.functional.interpolate(G_pred, (h, w), mode='bicubic', align_corners=False)
G_pred = G_pred[0, :].transpose([1, 2, 0])
G_pred = paddle.concat([G_pred, G_pred, G_pred], axis=-1)
G_pred = G_pred.detach().numpy()
......@@ -68,7 +67,7 @@ class SkyFilter():
frames_num = preview_frames_num if 0 < preview_frames_num < self.m_frames else self.m_frames
print('frames_num: %d, running evaluation...' % frames_num)
for idx in range(1, frames_num+1):
for idx in range(1, frames_num + 1):
ret, frame = self.cap.read()
if ret:
frame = cv2.resize(frame, self.out_size)
......@@ -84,7 +83,7 @@ class SkyFilter():
if self.is_show:
show_img = np.concatenate([frame, result], 1)
h, w = show_img.shape[:2]
show_img = cv2.resize(show_img, (720, int(720/w*h)))
show_img = cv2.resize(show_img, (720, int(720 / w * h)))
cv2.imshow('preview', show_img)
k = cv2.waitKey(1)
if (k == 27) or (cv2.getWindowProperty('preview', 0) == -1):
......
modules/thirdparty/video/Video_editing/SkyAR/utils.py
浏览文件 @ d6b6c6ff
......@@ -3,10 +3,7 @@ import numpy as np
from sklearn.neighbors import KernelDensity
__all__ = [
'build_transformation_matrix',
'update_transformation_matrix',
'estimate_partial_transform',
'removeOutliers',
'build_transformation_matrix', 'update_transformation_matrix', 'estimate_partial_transform', 'removeOutliers',
'guidedfilter'
]
......@@ -48,8 +45,7 @@ def estimate_partial_transform(matched_keypoints):
:return: transform as list of [dx, dy, da]
"""
prev_matched_kp, cur_matched_kp = matched_keypoints
transform = cv2.estimateAffinePartial2D(np.array(prev_matched_kp),
np.array(cur_matched_kp))[0]
transform = cv2.estimateAffinePartial2D(np.array(prev_matched_kp), np.array(cur_matched_kp))[0]
if transform is not None:
# translation x
......@@ -83,16 +79,14 @@ def boxfilter(img, r):
imDst = np.zeros_like(img)
imCum = np.cumsum(img, 0)
imDst[0: r+1, :] = imCum[r: 2*r+1, :]
imDst[r+1: rows-r, :] = imCum[2*r+1: rows, :] - imCum[0: rows-2*r-1, :]
imDst[rows-r: rows, :] = np.tile(imCum[rows-1, :],
[r, 1]) - imCum[rows-2*r-1: rows-r-1, :]
imDst[0:r + 1, :] = imCum[r:2 * r + 1, :]
imDst[r + 1:rows - r, :] = imCum[2 * r + 1:rows, :] - imCum[0:rows - 2 * r - 1, :]
imDst[rows - r:rows, :] = np.tile(imCum[rows - 1, :], [r, 1]) - imCum[rows - 2 * r - 1:rows - r - 1, :]
imCum = np.cumsum(imDst, 1)
imDst[:, 0: r+1] = imCum[:, r: 2*r+1]
imDst[:, r+1: cols-r] = imCum[:, 2*r+1: cols] - imCum[:, 0: cols-2*r-1]
imDst[:, cols-r: cols] = np.tile(imCum[:, cols-1],
[r, 1]).T - imCum[:, cols-2*r-1: cols-r-1]
imDst[:, 0:r + 1] = imCum[:, r:2 * r + 1]
imDst[:, r + 1:cols - r] = imCum[:, 2 * r + 1:cols] - imCum[:, 0:cols - 2 * r - 1]
imDst[:, cols - r:cols] = np.tile(imCum[:, cols - 1], [r, 1]).T - imCum[:, cols - 2 * r - 1:cols - r - 1]
return imDst
......
paddlehub/compat/module/module_desc_pb2.py
浏览文件 @ d6b6c6ff
......@@ -64,7 +64,8 @@ _KVDATA_KEYTYPEENTRY = _descriptor.Descriptor(
file=DESCRIPTOR,
containing_type=None,
fields=[
_descriptor.FieldDescriptor(name='key',
_descriptor.FieldDescriptor(
name='key',
full_name='paddlehub.module.desc.KVData.KeyTypeEntry.key',
index=0,
number=1,
......@@ -79,7 +80,8 @@ _KVDATA_KEYTYPEENTRY = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='value',
_descriptor.FieldDescriptor(
name='value',
full_name='paddlehub.module.desc.KVData.KeyTypeEntry.value',
index=1,
number=2,
......@@ -114,7 +116,8 @@ _KVDATA_DATAENTRY = _descriptor.Descriptor(
file=DESCRIPTOR,
containing_type=None,
fields=[
_descriptor.FieldDescriptor(name='key',
_descriptor.FieldDescriptor(
name='key',
full_name='paddlehub.module.desc.KVData.DataEntry.key',
index=0,
number=1,
......@@ -129,7 +132,8 @@ _KVDATA_DATAENTRY = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='value',
_descriptor.FieldDescriptor(
name='value',
full_name='paddlehub.module.desc.KVData.DataEntry.value',
index=1,
number=2,
......@@ -164,7 +168,8 @@ _KVDATA = _descriptor.Descriptor(
file=DESCRIPTOR,
containing_type=None,
fields=[
_descriptor.FieldDescriptor(name='key_type',
_descriptor.FieldDescriptor(
name='key_type',
full_name='paddlehub.module.desc.KVData.key_type',
index=0,
number=1,
......@@ -179,7 +184,8 @@ _KVDATA = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='data',
_descriptor.FieldDescriptor(
name='data',
full_name='paddlehub.module.desc.KVData.data',
index=1,
number=2,
......@@ -217,7 +223,8 @@ _MODULEATTR = _descriptor.Descriptor(
file=DESCRIPTOR,
containing_type=None,
fields=[
_descriptor.FieldDescriptor(name='type',
_descriptor.FieldDescriptor(
name='type',
full_name='paddlehub.module.desc.ModuleAttr.type',
index=0,
number=1,
......@@ -232,7 +239,8 @@ _MODULEATTR = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='i',
_descriptor.FieldDescriptor(
name='i',
full_name='paddlehub.module.desc.ModuleAttr.i',
index=1,
number=2,
......@@ -247,7 +255,8 @@ _MODULEATTR = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='f',
_descriptor.FieldDescriptor(
name='f',
full_name='paddlehub.module.desc.ModuleAttr.f',
index=2,
number=3,
......@@ -262,7 +271,8 @@ _MODULEATTR = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='b',
_descriptor.FieldDescriptor(
name='b',
full_name='paddlehub.module.desc.ModuleAttr.b',
index=3,
number=4,
......@@ -277,7 +287,8 @@ _MODULEATTR = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='s',
_descriptor.FieldDescriptor(
name='s',
full_name='paddlehub.module.desc.ModuleAttr.s',
index=4,
number=5,
......@@ -292,7 +303,8 @@ _MODULEATTR = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='map',
_descriptor.FieldDescriptor(
name='map',
full_name='paddlehub.module.desc.ModuleAttr.map',
index=5,
number=6,
......@@ -307,7 +319,8 @@ _MODULEATTR = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='list',
_descriptor.FieldDescriptor(
name='list',
full_name='paddlehub.module.desc.ModuleAttr.list',
index=6,
number=7,
......@@ -322,7 +335,8 @@ _MODULEATTR = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='set',
_descriptor.FieldDescriptor(
name='set',
full_name='paddlehub.module.desc.ModuleAttr.set',
index=7,
number=8,
......@@ -337,7 +351,8 @@ _MODULEATTR = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='object',
_descriptor.FieldDescriptor(
name='object',
full_name='paddlehub.module.desc.ModuleAttr.object',
index=8,
number=9,
......@@ -352,7 +367,8 @@ _MODULEATTR = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='name',
_descriptor.FieldDescriptor(
name='name',
full_name='paddlehub.module.desc.ModuleAttr.name',
index=9,
number=10,
......@@ -367,7 +383,8 @@ _MODULEATTR = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='info',
_descriptor.FieldDescriptor(
name='info',
full_name='paddlehub.module.desc.ModuleAttr.info',
index=10,
number=11,
......@@ -402,7 +419,8 @@ _FEEDDESC = _descriptor.Descriptor(
file=DESCRIPTOR,
containing_type=None,
fields=[
_descriptor.FieldDescriptor(name='var_name',
_descriptor.FieldDescriptor(
name='var_name',
full_name='paddlehub.module.desc.FeedDesc.var_name',
index=0,
number=1,
......@@ -417,7 +435,8 @@ _FEEDDESC = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='alias',
_descriptor.FieldDescriptor(
name='alias',
full_name='paddlehub.module.desc.FeedDesc.alias',
index=1,
number=2,
......@@ -452,7 +471,8 @@ _FETCHDESC = _descriptor.Descriptor(
file=DESCRIPTOR,
containing_type=None,
fields=[
_descriptor.FieldDescriptor(name='var_name',
_descriptor.FieldDescriptor(
name='var_name',
full_name='paddlehub.module.desc.FetchDesc.var_name',
index=0,
number=1,
......@@ -467,7 +487,8 @@ _FETCHDESC = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='alias',
_descriptor.FieldDescriptor(
name='alias',
full_name='paddlehub.module.desc.FetchDesc.alias',
index=1,
number=2,
......@@ -502,7 +523,8 @@ _MODULEVAR = _descriptor.Descriptor(
file=DESCRIPTOR,
containing_type=None,
fields=[
_descriptor.FieldDescriptor(name='fetch_desc',
_descriptor.FieldDescriptor(
name='fetch_desc',
full_name='paddlehub.module.desc.ModuleVar.fetch_desc',
index=0,
number=1,
......@@ -517,7 +539,8 @@ _MODULEVAR = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='feed_desc',
_descriptor.FieldDescriptor(
name='feed_desc',
full_name='paddlehub.module.desc.ModuleVar.feed_desc',
index=1,
number=2,
......@@ -552,7 +575,8 @@ _MODULEDESC_SIGN2VARENTRY = _descriptor.Descriptor(
file=DESCRIPTOR,
containing_type=None,
fields=[
_descriptor.FieldDescriptor(name='key',
_descriptor.FieldDescriptor(
name='key',
full_name='paddlehub.module.desc.ModuleDesc.Sign2varEntry.key',
index=0,
number=1,
......@@ -567,7 +591,8 @@ _MODULEDESC_SIGN2VARENTRY = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='value',
_descriptor.FieldDescriptor(
name='value',
full_name='paddlehub.module.desc.ModuleDesc.Sign2varEntry.value',
index=1,
number=2,
......@@ -602,7 +627,8 @@ _MODULEDESC = _descriptor.Descriptor(
file=DESCRIPTOR,
containing_type=None,
fields=[
_descriptor.FieldDescriptor(name='sign2var',
_descriptor.FieldDescriptor(
name='sign2var',
full_name='paddlehub.module.desc.ModuleDesc.sign2var',
index=0,
number=2,
......@@ -617,7 +643,8 @@ _MODULEDESC = _descriptor.Descriptor(
is_extension=False,
extension_scope=None,
options=None),
_descriptor.FieldDescriptor(name='attr',
_descriptor.FieldDescriptor(
name='attr',
full_name='paddlehub.module.desc.ModuleDesc.attr',
index=1,
number=3,
......@@ -679,14 +706,16 @@ KVData = _reflection.GeneratedProtocolMessageType(
KeyTypeEntry=_reflection.GeneratedProtocolMessageType(
'KeyTypeEntry',
(_message.Message, ),
dict(DESCRIPTOR=_KVDATA_KEYTYPEENTRY,
dict(
DESCRIPTOR=_KVDATA_KEYTYPEENTRY,
__module__='module_desc_pb2'
# @@protoc_insertion_point(class_scope:paddlehub.module.desc.KVData.KeyTypeEntry)
)),
DataEntry=_reflection.GeneratedProtocolMessageType(
'DataEntry',
(_message.Message, ),
dict(DESCRIPTOR=_KVDATA_DATAENTRY,
dict(
DESCRIPTOR=_KVDATA_DATAENTRY,
__module__='module_desc_pb2'
# @@protoc_insertion_point(class_scope:paddlehub.module.desc.KVData.DataEntry)
)),
......@@ -701,7 +730,8 @@ _sym_db.RegisterMessage(KVData.DataEntry)
ModuleAttr = _reflection.GeneratedProtocolMessageType(
'ModuleAttr',
(_message.Message, ),
dict(DESCRIPTOR=_MODULEATTR,
dict(
DESCRIPTOR=_MODULEATTR,
__module__='module_desc_pb2'
# @@protoc_insertion_point(class_scope:paddlehub.module.desc.ModuleAttr)
))
......@@ -710,7 +740,8 @@ _sym_db.RegisterMessage(ModuleAttr)
FeedDesc = _reflection.GeneratedProtocolMessageType(
'FeedDesc',
(_message.Message, ),
dict(DESCRIPTOR=_FEEDDESC,
dict(
DESCRIPTOR=_FEEDDESC,
__module__='module_desc_pb2'
# @@protoc_insertion_point(class_scope:paddlehub.module.desc.FeedDesc)
))
......@@ -719,7 +750,8 @@ _sym_db.RegisterMessage(FeedDesc)
FetchDesc = _reflection.GeneratedProtocolMessageType(
'FetchDesc',
(_message.Message, ),
dict(DESCRIPTOR=_FETCHDESC,
dict(
DESCRIPTOR=_FETCHDESC,
__module__='module_desc_pb2'
# @@protoc_insertion_point(class_scope:paddlehub.module.desc.FetchDesc)
))
......@@ -728,7 +760,8 @@ _sym_db.RegisterMessage(FetchDesc)
ModuleVar = _reflection.GeneratedProtocolMessageType(
'ModuleVar',
(_message.Message, ),
dict(DESCRIPTOR=_MODULEVAR,
dict(
DESCRIPTOR=_MODULEVAR,
__module__='module_desc_pb2'
# @@protoc_insertion_point(class_scope:paddlehub.module.desc.ModuleVar)
))
......@@ -741,7 +774,8 @@ ModuleDesc = _reflection.GeneratedProtocolMessageType(
Sign2varEntry=_reflection.GeneratedProtocolMessageType(
'Sign2varEntry',
(_message.Message, ),
dict(DESCRIPTOR=_MODULEDESC_SIGN2VARENTRY,
dict(
DESCRIPTOR=_MODULEDESC_SIGN2VARENTRY,
__module__='module_desc_pb2'
# @@protoc_insertion_point(class_scope:paddlehub.module.desc.ModuleDesc.Sign2varEntry)
)),
......
paddlehub/datasets/base_audio_dataset.py
浏览文件 @ d6b6c6ff
......@@ -27,6 +27,7 @@ class InputExample(object):
"""
Input example of one audio sample.
"""
def __init__(self, guid: int, source: Union[list, str], label: Optional[str] = None):
self.guid = guid
self.source = source
......@@ -37,6 +38,7 @@ class BaseAudioDataset(object):
"""
Base class of speech dataset.
"""
def __init__(self, base_path: str, data_file: str, mode: Optional[str] = "train"):
self.data_file = os.path.join(base_path, data_file)
self.mode = mode
......@@ -92,7 +94,8 @@ class AudioClassificationDataset(BaseAudioDataset, paddle.io.Dataset):
if self.feat_type == 'raw':
record['feat'] = example.source
elif self.feat_type == 'mel':
record['feat'] = extract_melspectrogram(example.source,
record['feat'] = extract_melspectrogram(
example.source,
sample_rate=self.feat_cfg['sample_rate'],
window_size=self.feat_cfg['window_size'],
hop_size=self.feat_cfg['hop_size'],
......
paddlehub/datasets/base_nlp_dataset.py
浏览文件 @ d6b6c6ff
......@@ -557,7 +557,9 @@ class TextMatchingDataset(BaseNLPDataset, paddle.io.Dataset):
pad_to_max_seq_len=True, return_length=True)
record = {'text_a': record_a, 'text_b': record_b}
else:
raise RuntimeError("Unknown type of self.tokenizer: {}, it must be an instance of PretrainedTokenizer".format(type(self.tokenizer)))
raise RuntimeError(
"Unknown type of self.tokenizer: {}, it must be an instance of PretrainedTokenizer".format(
type(self.tokenizer)))
if not record:
logger.info(
......@@ -582,7 +584,9 @@ class TextMatchingDataset(BaseNLPDataset, paddle.io.Dataset):
else:
return query_input_ids, query_token_type_ids, title_input_ids, title_token_type_ids
else:
raise RuntimeError("Unknown type of self.tokenizer: {}, it must be an instance of PretrainedTokenizer".format(type(self.tokenizer)))
raise RuntimeError(
"Unknown type of self.tokenizer: {}, it must be an instance of PretrainedTokenizer".format(
type(self.tokenizer)))
def __len__(self):
return len(self.records)
paddlehub/datasets/base_seg_dataset.py
浏览文件 @ d6b6c6ff
......@@ -63,46 +63,34 @@ class SegDataset(paddle.io.Dataset):
self.edge = edge
if mode.lower() not in ['train', 'val', 'test']:
raise ValueError(
"mode should be 'train', 'val' or 'test', but got {}.".format(
mode))
raise ValueError("mode should be 'train', 'val' or 'test', but got {}.".format(mode))
if self.transforms is None:
raise ValueError("`transforms` is necessary, but it is None.")
self.dataset_root = dataset_root
if not os.path.exists(self.dataset_root):
raise FileNotFoundError('there is not `dataset_root`: {}.'.format(
self.dataset_root))
raise FileNotFoundError('there is not `dataset_root`: {}.'.format(self.dataset_root))
if mode == 'train':
if train_path is None:
raise ValueError(
'When `mode` is "train", `train_path` is necessary, but it is None.'
)
raise ValueError('When `mode` is "train", `train_path` is necessary, but it is None.')
elif not os.path.exists(train_path):
raise FileNotFoundError(
'`train_path` is not found: {}'.format(train_path))
raise FileNotFoundError('`train_path` is not found: {}'.format(train_path))
else:
file_path = train_path
elif mode == 'val':
if val_path is None:
raise ValueError(
'When `mode` is "val", `val_path` is necessary, but it is None.'
)
raise ValueError('When `mode` is "val", `val_path` is necessary, but it is None.')
elif not os.path.exists(val_path):
raise FileNotFoundError(
'`val_path` is not found: {}'.format(val_path))
raise FileNotFoundError('`val_path` is not found: {}'.format(val_path))
else:
file_path = val_path
else:
if test_path is None:
raise ValueError(
'When `mode` is "test", `test_path` is necessary, but it is None.'
)
raise ValueError('When `mode` is "test", `test_path` is necessary, but it is None.')
elif not os.path.exists(test_path):
raise FileNotFoundError(
'`test_path` is not found: {}'.format(test_path))
raise FileNotFoundError('`test_path` is not found: {}'.format(test_path))
else:
file_path = test_path
......@@ -111,8 +99,7 @@ class SegDataset(paddle.io.Dataset):
items = line.strip().split(separator)
if len(items) != 2:
if mode == 'train' or mode == 'val':
raise ValueError(
"File list format incorrect! In training or evaluation task it should be"
raise ValueError("File list format incorrect! In training or evaluation task it should be"
" image_name{}label_name\\n".format(separator))
image_path = os.path.join(self.dataset_root, items[0])
label_path = None
......@@ -138,4 +125,3 @@ class SegDataset(paddle.io.Dataset):
def __len__(self) -> int:
return len(self.file_list)
paddlehub/datasets/esc50.py
浏览文件 @ d6b6c6ff
......@@ -91,7 +91,8 @@ class ESC50(AudioClassificationDataset):
else:
data_file = 'dev.npz'
feat_cfg = dict(sample_rate=self.sample_rate,
feat_cfg = dict(
sample_rate=self.sample_rate,
window_size=1024,
hop_size=320,
mel_bins=64,
......@@ -99,7 +100,8 @@ class ESC50(AudioClassificationDataset):
fmax=14000,
window='hann')
super().__init__(base_path=base_path,
super().__init__(
base_path=base_path,
data_file=data_file,
file_type='npz',
mode=mode,
......
paddlehub/datasets/opticdiscseg.py
浏览文件 @ d6b6c6ff
......@@ -24,6 +24,7 @@ import paddlehub.env as hubenv
from paddlehub.utils.download import download_data
from paddlehub.datasets.base_seg_dataset import SegDataset
@download_data(url='https://paddleseg.bj.bcebos.com/dataset/optic_disc_seg.zip')
class OpticDiscSeg(SegDataset):
"""
......@@ -36,9 +37,7 @@ class OpticDiscSeg(SegDataset):
edge (bool, optional): Whether to compute edge while training. Default: False
"""
def __init__(self,
transforms: Callable = None,
mode: str = 'train'):
def __init__(self, transforms: Callable = None, mode: str = 'train'):
self.transforms = transforms
mode = mode.lower()
self.mode = mode
......@@ -47,14 +46,11 @@ class OpticDiscSeg(SegDataset):
self.ignore_index = 255
if mode not in ['train', 'val', 'test']:
raise ValueError(
"`mode` should be 'train', 'val' or 'test', but got {}.".format(
mode))
raise ValueError("`mode` should be 'train', 'val' or 'test', but got {}.".format(mode))
if self.transforms is None:
raise ValueError("`transforms` is necessary, but it is None.")
if mode == 'train':
file_path = os.path.join(hubenv.DATA_HOME, 'optic_disc_seg', 'train_list.txt')
elif mode == 'test':
......@@ -67,9 +63,7 @@ class OpticDiscSeg(SegDataset):
items = line.strip().split()
if len(items) != 2:
if mode == 'train' or mode == 'val':
raise Exception(
"File list format incorrect! It should be"
" image_name label_name\\n")
raise Exception("File list format incorrect! It should be" " image_name label_name\\n")
image_path = os.path.join(hubenv.DATA_HOME, 'optic_disc_seg', items[0])
grt_path = None
else:
......
paddlehub/datasets/pascalvoc.py
浏览文件 @ d6b6c6ff
......@@ -35,6 +35,7 @@ class DetectCatagory:
label_ids(List(int)): The dataset label ids.
category_to_id_map(dict): Mapping relations of category and id for images.
"""
def __init__(self, attrbox: Callable, data_dir: str):
self.attrbox = attrbox
self.img_dir = data_dir
......@@ -62,6 +63,7 @@ class ParseImages:
Returns:
imgs(dict): The input for detection model, it is a dict.
"""
def __init__(self, attrbox: Callable, data_dir: str, category_to_id_map: dict):
self.attrbox = attrbox
self.img_dir = data_dir
......@@ -94,6 +96,7 @@ class GTAnotations:
Returns:
img(dict): Set specific value on the attributes of 'gt boxes' and 'gt labels' for input.
"""
def __init__(self, attrbox: Callable, category_to_id_map: dict):
self.attrbox = attrbox
self.category_to_id_map = category_to_id_map
......@@ -154,6 +157,7 @@ class DetectionData(paddle.io.Dataset):
Returns:
DataSet: An iterable object for data iterating
"""
def __init__(self, transform: Callable, size: int = 416, mode: str = 'train'):
self.mode = mode
self.transform = transform
......
paddlehub/finetune/trainer.py
浏览文件 @ d6b6c6ff
......@@ -68,7 +68,6 @@ class Trainer(object):
if not isinstance(self.model, paddle.nn.Layer):
raise TypeError('The model {} is not a `paddle.nn.Layer` object.'.format(self.model.__name__))
if self.local_rank == 0 and not os.path.exists(self.checkpoint_dir):
os.makedirs(self.checkpoint_dir)
......
paddlehub/module/cv_module.py
浏览文件 @ d6b6c6ff
......@@ -126,8 +126,8 @@ class ImageClassifierModule(RunModule, ImageServing):
"""
top_k = int(top_k)
images_decode = [base64_to_cv2(image) for image in images]
resdicts = self.predict(images=images_decode, top_k=top_k,**kwargs)
final={}
resdicts = self.predict(images=images_decode, top_k=top_k, **kwargs)
final = {}
for resdict in resdicts:
for key, value in resdict.items():
resdict[key] = float(value)
......@@ -144,18 +144,13 @@ class ImageClassifierModule(RunModule, ImageServing):
prog='hub run {}'.format(self.name),
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, not required.")
title="Config options", description="Run configuration for controlling module behavior, not required.")
self.add_module_config_arg()
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
results = self.predict(
images=[args.input_path],
top_k=args.top_k)
results = self.predict(images=[args.input_path], top_k=args.top_k)
return results
......@@ -164,18 +159,13 @@ class ImageClassifierModule(RunModule, ImageServing):
Add the command config options.
"""
self.arg_config_group.add_argument(
'--top_k',
type=int,
default=1,
help="top_k classification result.")
self.arg_config_group.add_argument('--top_k', type=int, default=1, help="top_k classification result.")
def add_module_input_arg(self):
"""
Add the command input options.
"""
self.arg_input_group.add_argument(
'--input_path', type=str, help="path to image.")
self.arg_input_group.add_argument('--input_path', type=str, help="path to image.")
class ImageColorizeModule(RunModule, ImageServing):
......@@ -288,11 +278,11 @@ class ImageColorizeModule(RunModule, ImageServing):
"""
images_decode = [base64_to_cv2(image) for image in images]
visual = self.predict(images=images_decode, **kwargs)
final={}
final = {}
for i, visual_ret in enumerate(visual):
h, w, c = images_decode[i].shape
for key, value in visual_ret.items():
value = cv2.resize(cv2.cvtColor(value,cv2.COLOR_RGB2BGR), (w, h), cv2.INTER_NEAREST)
value = cv2.resize(cv2.cvtColor(value, cv2.COLOR_RGB2BGR), (w, h), cv2.INTER_NEAREST)
visual_ret[key] = cv2_to_base64(value)
final['data'] = visual
return final
......@@ -307,19 +297,13 @@ class ImageColorizeModule(RunModule, ImageServing):
prog='hub run {}'.format(self.name),
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, not required.")
title="Config options", description="Run configuration for controlling module behavior, not required.")
self.add_module_config_arg()
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
results = self.predict(
images=[args.input_path],
visualization=args.visualization,
save_path=args.output_dir)
results = self.predict(images=[args.input_path], visualization=args.visualization, save_path=args.output_dir)
return results
......@@ -329,22 +313,15 @@ class ImageColorizeModule(RunModule, ImageServing):
"""
self.arg_config_group.add_argument(
'--output_dir',
type=str,
default='colorization',
help="save visualization result.")
'--output_dir', type=str, default='colorization', help="save visualization result.")
self.arg_config_group.add_argument(
'--visualization',
type=bool,
default=True,
help="whether to save output as images.")
'--visualization', type=bool, default=True, help="whether to save output as images.")
def add_module_input_arg(self):
"""
Add the command input options.
"""
self.arg_input_group.add_argument(
'--input_path', type=str, help="path to image.")
self.arg_input_group.add_argument('--input_path', type=str, help="path to image.")
class Yolov3Module(RunModule, ImageServing):
......@@ -523,7 +500,12 @@ class StyleTransferModule(RunModule, ImageServing):
return {'loss': loss, 'metrics': {'content gap': content_loss, 'style gap': style_loss}}
def predict(self, origin: list, style: Union[str, np.ndarray], batch_size: int = 1, visualization: bool = True, save_path: str = 'style_tranfer'):
def predict(self,
origin: list,
style: Union[str, np.ndarray],
batch_size: int = 1,
visualization: bool = True,
save_path: str = 'style_tranfer'):
'''
Colorize images
......@@ -578,7 +560,7 @@ class StyleTransferModule(RunModule, ImageServing):
images_decode = [base64_to_cv2(image) for image in images[0]]
style_decode = base64_to_cv2(images[1])
results = self.predict(origin=images_decode, style=style_decode, **kwargs)
final={}
final = {}
final['data'] = [cv2_to_base64(result) for result in results]
return final
......@@ -592,12 +574,9 @@ class StyleTransferModule(RunModule, ImageServing):
prog='hub run {}'.format(self.name),
usage='%(prog)s',
add_help=True)
self.arg_input_group = self.parser.add_argument_group(
title="Input options", description="Input data. Required")
self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
self.arg_config_group = self.parser.add_argument_group(
title="Config options",
description=
"Run configuration for controlling module behavior, not required.")
title="Config options", description="Run configuration for controlling module behavior, not required.")
self.add_module_config_arg()
self.add_module_input_arg()
args = self.parser.parse_args(argvs)
......@@ -615,25 +594,17 @@ class StyleTransferModule(RunModule, ImageServing):
"""
self.arg_config_group.add_argument(
'--output_dir',
type=str,
default='style_tranfer',
help="The directory to save output images.")
'--output_dir', type=str, default='style_tranfer', help="The directory to save output images.")
self.arg_config_group.add_argument(
'--visualization',
type=bool,
default=True,
help="whether to save output as images.")
'--visualization', type=bool, default=True, help="whether to save output as images.")
def add_module_input_arg(self):
"""
Add the command input options.
"""
self.arg_input_group.add_argument(
'--input_path', type=str, help="path to image.")
self.arg_input_group.add_argument(
'--style_path', type=str, help="path to style image.")
self.arg_input_group.add_argument('--input_path', type=str, help="path to image.")
self.arg_input_group.add_argument('--style_path', type=str, help="path to style image.")
class ImageSegmentationModule(ImageServing, RunModule):
......@@ -659,13 +630,17 @@ class ImageSegmentationModule(ImageServing, RunModule):
if logit.shape[-2:] != label.shape[-2:]:
logit = F.interpolate(logit, label.shape[-2:], mode='bilinear')
logit = logit.transpose([0,2,3,1])
logit = logit.transpose([0, 2, 3, 1])
loss_ce = criterionCE(logit, label)
loss += loss_ce / len(logits)
return {"loss": loss}
def predict(self, images: Union[str, np.ndarray], batch_size: int = 1, visualization: bool = True, save_path: str = 'seg_result') -> List[np.ndarray]:
def predict(self,
images: Union[str, np.ndarray],
batch_size: int = 1,
visualization: bool = True,
save_path: str = 'seg_result') -> List[np.ndarray]:
'''
Obtain segmentation results.
......@@ -679,7 +654,7 @@ class ImageSegmentationModule(ImageServing, RunModule):
output(list[np.ndarray]) : The segmentation mask.
'''
self.eval()
result=[]
result = []
total_num = len(images)
loop_num = int(np.ceil(total_num / batch_size))
......@@ -697,17 +672,17 @@ class ImageSegmentationModule(ImageServing, RunModule):
pred = paddle.argmax(pred[0], axis=1, keepdim=True, dtype='int32')
for num in range(pred.shape[0]):
if isinstance(images[handle_id+num], str):
image = cv2.imread(images[handle_id+num])
if isinstance(images[handle_id + num], str):
image = cv2.imread(images[handle_id + num])
else:
image = images[handle_id+num]
image = images[handle_id + num]
h, w, c = image.shape
pred_final = utils.reverse_transform(pred[num: num+1], (h,w), self.transforms.transforms)
pred_final = utils.reverse_transform(pred[num:num + 1], (h, w), self.transforms.transforms)
pred_final = paddle.squeeze(pred_final)
pred_final = pred_final.numpy().astype('uint8')
if visualization:
added_image = utils.visualize(images[handle_id+num], pred_final, weight=0.6)
added_image = utils.visualize(images[handle_id + num], pred_final, weight=0.6)
pred_mask = utils.get_pseudo_color_map(pred_final)
pred_image_path = os.path.join(save_path, 'image', str(time.time()) + ".png")
pred_mask_path = os.path.join(save_path, 'mask', str(time.time()) + ".png")
......@@ -728,7 +703,7 @@ class ImageSegmentationModule(ImageServing, RunModule):
"""
images_decode = [base64_to_cv2(image) for image in images]
visual = self.predict(images=images_decode, **kwargs)
final=[]
final = []
for mask in visual:
final.append(cv2_to_base64(mask))
return final
paddlehub/utils/download.py
浏览文件 @ d6b6c6ff
......@@ -23,7 +23,6 @@ from paddlehub.utils import log, utils, xarfile
def download_data(url):
def _wrapper(Dataset):
def _check_download():
save_name = os.path.basename(url).split('.')[0]
......
paddlehub/utils/utils.py
浏览文件 @ d6b6c6ff
......@@ -40,6 +40,7 @@ from paddlehub.utils.log import logger
class Version(packaging.version.Version):
'''Extended implementation of packaging.version.Version'''
def match(self, condition: str) -> bool:
'''
Determine whether the given condition are met
......@@ -105,6 +106,7 @@ class Version(packaging.version.Version):
class Timer(object):
'''Calculate runing speed and estimated time of arrival(ETA)'''
def __init__(self, total_step: int):
self.total_step = total_step
self.last_start_step = 0
......@@ -408,7 +410,8 @@ def extract_melspectrogram(y,
logger.error('Failed to import librosa. Please check that librosa and numba are correctly installed.')
raise
s = librosa.stft(y,
s = librosa.stft(
y,
n_fft=window_size,
hop_length=hop_size,
win_length=window_size,
......
paddlehub/vision/segmentation_transforms.py
浏览文件 @ d6b6c6ff
......@@ -145,18 +145,15 @@ class Padding:
label_padding_value: int = 255):
if isinstance(target_size, list) or isinstance(target_size, tuple):
if len(target_size) != 2:
raise ValueError(
'`target_size` should include 2 elements, but it is {}'.
format(target_size))
raise ValueError('`target_size` should include 2 elements, but it is {}'.format(target_size))
else:
raise TypeError(
"Type of target_size is invalid. It should be list or tuple, now is {}"
.format(type(target_size)))
raise TypeError("Type of target_size is invalid. It should be list or tuple, now is {}".format(
type(target_size)))
self.target_size = target_size
self.im_padding_value = im_padding_value
self.label_padding_value = label_padding_value
def __call__(self, im: np.ndarray , label: np.ndarray = None) -> Tuple:
def __call__(self, im: np.ndarray, label: np.ndarray = None) -> Tuple:
"""
Args:
im (np.ndarray): The Image data.
......@@ -179,13 +176,12 @@ class Padding:
'The size of image should be less than `target_size`, but the size of image ({}, {}) is larger than `target_size` ({}, {})'
.format(im_width, im_height, target_width, target_height))
else:
im = cv2.copyMakeBorder(im, 0, pad_height, 0, pad_width, cv2.BORDER_CONSTANT,
value=self.im_padding_value)
im = cv2.copyMakeBorder(im, 0, pad_height, 0, pad_width, cv2.BORDER_CONSTANT, value=self.im_padding_value)
if label is not None:
label = cv2.copyMakeBorder(label, 0, pad_height, 0, pad_width, cv2.BORDER_CONSTANT,
value=self.label_padding_value)
label = cv2.copyMakeBorder(
label, 0, pad_height, 0, pad_width, cv2.BORDER_CONSTANT, value=self.label_padding_value)
if label is None:
return (im,)
return (im, )
else:
return (im, label)
......@@ -200,15 +196,13 @@ class Normalize:
ValueError: When mean/std is not list or any value in std is 0.
"""
def __init__(self, mean: Union[List[float], Tuple[float]] = (0.5, 0.5, 0.5),
def __init__(self,
mean: Union[List[float], Tuple[float]] = (0.5, 0.5, 0.5),
std: Union[List[float], Tuple[float]] = (0.5, 0.5, 0.5)):
self.mean = mean
self.std = std
if not (isinstance(self.mean, (list, tuple))
and isinstance(self.std, (list, tuple))):
raise ValueError(
"{}: input type is invalid. It should be list or tuple".format(
self))
if not (isinstance(self.mean, (list, tuple)) and isinstance(self.std, (list, tuple))):
raise ValueError("{}: input type is invalid. It should be list or tuple".format(self))
from functools import reduce
if reduce(lambda x, y: x * y, self.std) == 0:
raise ValueError('{}: std is invalid!'.format(self))
......@@ -227,7 +221,7 @@ class Normalize:
im = F.normalize(im, mean, std)
if label is None:
return (im,)
return (im, )
else:
return (im, label)
......@@ -260,17 +254,13 @@ class Resize:
def __init__(self, target_size: Union[List[int], Tuple[int]] = (512, 512), interp: str = 'LINEAR'):
self.interp = interp
if not (interp == "RANDOM" or interp in self.interp_dict):
raise ValueError("`interp` should be one of {}".format(
self.interp_dict.keys()))
raise ValueError("`interp` should be one of {}".format(self.interp_dict.keys()))
if isinstance(target_size, list) or isinstance(target_size, tuple):
if len(target_size) != 2:
raise ValueError(
'`target_size` should include 2 elements, but it is {}'.
format(target_size))
raise ValueError('`target_size` should include 2 elements, but it is {}'.format(target_size))
else:
raise TypeError(
"Type of `target_size` is invalid. It should be list or tuple, but it is {}"
.format(type(target_size)))
raise TypeError("Type of `target_size` is invalid. It should be list or tuple, but it is {}".format(
type(target_size)))
self.target_size = target_size
......@@ -298,10 +288,9 @@ class Resize:
interp = self.interp
im = F.resize(im, self.target_size, self.interp_dict[interp])
if label is not None:
label = F.resize(label, self.target_size,
cv2.INTER_NEAREST)
label = F.resize(label, self.target_size, cv2.INTER_NEAREST)
if label is None:
return (im,)
return (im, )
else:
return (im, label)
paddlehub/vision/transforms.py
浏览文件 @ d6b6c6ff
......@@ -31,6 +31,7 @@ class Compose:
to_rgb(bool): Whether to transform the input from BGR mode to RGB mode, default is False.
channel_first(bool): whether to permute image from channel laste to channel first
"""
def __init__(self, transforms: Callable, to_rgb: bool = False, channel_first: bool = True):
if not isinstance(transforms, list):
raise TypeError('The transforms must be a list!')
......@@ -57,16 +58,20 @@ class Compose:
im = F.permute(im)
return im
class Permute:
"""
Repermute the input image from [H, W, C] to [C, H, W].
"""
def __init__(self):
pass
def __call__(self, im):
im = F.permute(im)
return im
class RandomHorizontalFlip:
"""
Randomly flip the image horizontally according to given probability.
......@@ -74,6 +79,7 @@ class RandomHorizontalFlip:
Args:
prob(float): The probability for flipping the image horizontally, default is 0.5.
"""
def __init__(self, prob: float = 0.5):
self.prob = prob
......@@ -90,6 +96,7 @@ class RandomVerticalFlip:
Args:
prob(float): The probability for flipping the image vertically, default is 0.5.
"""
def __init__(self, prob: float = 0.5):
self.prob = prob
......@@ -146,6 +153,7 @@ class ResizeByLong:
Args:
long_size(int|list[int]): The target size of long side.
"""
def __init__(self, long_size: Union[List[int], int]):
self.long_size = long_size
......@@ -162,6 +170,7 @@ class ResizeRangeScaling:
min_value(int): The minimum value for targeted size.
max_value(int): The maximum value for targeted size.
"""
def __init__(self, min_value: int = 400, max_value: int = 600):
if min_value > max_value:
raise ValueError('min_value must be less than max_value, '
......@@ -188,6 +197,7 @@ class ResizeStepScaling:
scale_step_size(float): Scale interval.
"""
def __init__(self, min_scale_factor: float = 0.75, max_scale_factor: float = 1.25, scale_step_size: float = 0.25):
if min_scale_factor > max_scale_factor:
raise ValueError('min_scale_factor must be less than max_scale_factor, '
......@@ -224,6 +234,7 @@ class Normalize:
std(list): Standard deviation for normalization.
channel_first(bool): im channel firest or last
"""
def __init__(self, mean: list = [0.5, 0.5, 0.5], std: list = [0.5, 0.5, 0.5], channel_first: bool = False):
self.mean = mean
self.std = std
......@@ -253,6 +264,7 @@ class Padding:
target_size(Union[List[int], Tuple[int], int]): Targeted image size.
im_padding_value(list): Border value for 3 channels, default is [127.5, 127.5, 127.5].
"""
def __init__(self, target_size: Union[List[int], Tuple[int], int], im_padding_value: list = [127.5, 127.5, 127.5]):
if isinstance(target_size, list) or isinstance(target_size, tuple):
if len(target_size) != 2:
......@@ -292,6 +304,7 @@ class RandomPaddingCrop:
crop_size(Union[List[int], Tuple[int], int]): Targeted image size.
im_padding_value(list): Border value for 3 channels, default is [127.5, 127.5, 127.5].
"""
def __init__(self, crop_size, im_padding_value=[127.5, 127.5, 127.5]):
if isinstance(crop_size, list) or isinstance(crop_size, tuple):
if len(crop_size) != 2:
......@@ -339,6 +352,7 @@ class RandomBlur:
Args:
prob(float): The probability to blur the image, default is 0.1.
"""
def __init__(self, prob: float = 0.1):
self.prob = prob
......@@ -370,6 +384,7 @@ class RandomRotation:
max_rotation(float): Upper bound of rotation angle.
im_padding_value(list): Border value for 3 channels, default is [127.5, 127.5, 127.5].
"""
def __init__(self, max_rotation: float = 15, im_padding_value: list = [127.5, 127.5, 127.5]):
self.max_rotation = max_rotation
self.im_padding_value = im_padding_value
......@@ -401,8 +416,6 @@ class RandomRotation:
return im
class RandomDistort:
"""
Random adjust brightness, contrast, saturation and hue according to the given random range and probability, respectively.
......@@ -418,6 +431,7 @@ class RandomDistort:
hue_range(float): Boundary of hue.
hue_prob(float): Probability for disturb the hue of image.
"""
def __init__(self,
brightness_range: float = 0.5,
brightness_prob: float = 0.5,
......
tests/test_module.py
浏览文件 @ d6b6c6ff
......@@ -14,10 +14,8 @@ class TestHubModule(unittest.TestCase):
self.assertEqual(results[0]['tag'], ['TIME', 'v', 'q', 'n'])
self.assertEqual(results[1]['word'], ['天气预报', '说', '今天', '要', '下雨'])
self.assertEqual(results[1]['tag'], ['n', 'v', 'TIME', 'v', 'v'])
self.assertEqual(results[2]['word'],
['下', '一班', '地铁', '马上', '就要', '到', '了'])
self.assertEqual(results[2]['tag'],
['f', 'm', 'n', 'd', 'v', 'v', 'xc'])
self.assertEqual(results[2]['word'], ['下', '一班', '地铁', '马上', '就要', '到', '了'])
self.assertEqual(results[2]['tag'], ['f', 'm', 'n', 'd', 'v', 'v', 'xc'])
def test_senta(self):
senta = hub.Module(name="senta_bilstm")
......@@ -45,13 +43,11 @@ class TestHubModule(unittest.TestCase):
if result['similarity'] > max_score:
max_score = result['similarity']
result_text = result['text_2']
print("The most matching with the %s is %s" % (test_text_1[0],
result_text))
print("The most matching with the %s is %s" % (test_text_1[0], result_text))
def test_ssd(self):
ssd = hub.Module(name="ssd_mobilenet_v1_pascal")
test_img_path = os.path.join(
os.path.dirname(__file__), "resources", "test_img_cat.jpg")
test_img_path = os.path.join(os.path.dirname(__file__), "resources", "test_img_cat.jpg")
input_dict = {"image": [test_img_path]}
results = ssd.object_detection(data=input_dict)
for result in results:
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或