Skip to content

P
PaddleDetection

s920243400 / PaddleDetection
与 Fork 源项目一致

Fork自 PaddlePaddle / PaddleDetection

通知 2
Star 0
Fork 0
P
PaddleDetection

体验新版 GitCode，发现更多精彩内容 >>

提交 9a17d02c 编写于 作者: W whs 提交者： GitHub
浏览文件
操作

[PaddleSlim] Add pruning demo for yolov3 (#3414)

上级 fe350028
隐藏空白更改
内联 并排
Showing with 480 addition and 0 deletion
slim/prune/README.md 0 → 100644
浏览文件 @ 9a17d02c
>运行该示例前请安装Paddle1.6或更高版本
# 检测模型卷积通道剪裁示例
## 概述
该示例使用PaddleSlim提供的[卷积通道剪裁压缩策略](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/docs/tutorial.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)对检测库中的模型进行压缩。
在阅读该示例前，建议您先了解以下内容：
- <a href="../..README_cn.md">检测库的常规训练方法</a>
- [PaddleSlim使用文档](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/docs/usage.md)
## 配置文件说明
关于配置文件如何编写您可以参考：
- [PaddleSlim配置文件编写说明](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/docs/usage.md#122-%E9%85%8D%E7%BD%AE%E6%96%87%E4%BB%B6%E7%9A%84%E4%BD%BF%E7%94%A8)
- [裁剪策略配置文件编写说明](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/docs/usage.md#22-%E6%A8%A1%E5%9E%8B%E9%80%9A%E9%81%93%E5%89%AA%E8%A3%81)
其中，配置文件中的`pruned_params`需要根据当前模型的网络结构特点设置，它用来指定要裁剪的parameters.
这里以MobileNetV1-YoloV3模型为例，其卷积可以三种：主干网络中的普通卷积，主干网络中的`depthwise convolution`和`yolo block`里的普通卷积。PaddleSlim暂时无法对`depthwise convolution`直接进行剪裁， 因为`depthwise convolution`的`channel`的变化会同时影响到前后的卷积层。我们这里只对主干网络中的普通卷积和`yolo block`里的普通卷积做裁剪。
通过以下方式可视化模型结构：
```
from paddle.fluid.framework import IrGraph
from paddle.fluid import core
graph = IrGraph(core.Graph(train_prog.desc), for_test=True)
marked_nodes = set()
for op in graph.all_op_nodes():
print(op.name())
if op.name().find('conv') > -1:
marked_nodes.add(op)
graph.draw('.', 'forward', marked_nodes)
```
该示例中MobileNetV1-YoloV3模型结构的可视化结果：<a href="./images/MobileNetV1-YoloV3.pdf">MobileNetV1-YoloV3.pdf</a>
同时通过以下命令观察目标卷积层的参数（parameters）的名称和shape:
```
for param in fluid.default_main_program().global_block().all_parameters():
if 'weights' in param.name:
print param.name, param.shape
```
从可视化结果，我们可以排除后续会做concat的卷积层，最终得到如下要裁剪的参数名称：
```
conv2_1_sep_weights
conv2_2_sep_weights
conv3_1_sep_weights
conv4_1_sep_weights
conv5_1_sep_weights
conv5_2_sep_weights
conv5_3_sep_weights
conv5_4_sep_weights
conv5_5_sep_weights
conv5_6_sep_weights
yolo_block.0.0.0.conv.weights
yolo_block.0.0.1.conv.weights
yolo_block.0.1.0.conv.weights
yolo_block.0.1.1.conv.weights
yolo_block.1.0.0.conv.weights
yolo_block.1.0.1.conv.weights
yolo_block.1.1.0.conv.weights
yolo_block.1.1.1.conv.weights
yolo_block.1.2.conv.weights
yolo_block.2.0.0.conv.weights
yolo_block.2.0.1.conv.weights
yolo_block.2.1.1.conv.weights
yolo_block.2.2.conv.weights
yolo_block.2.tip.conv.weights
```
```
(conv2_1_sep_weights)|(conv2_2_sep_weights)|(conv3_1_sep_weights)|(conv4_1_sep_weights)|(conv5_1_sep_weights)|(conv5_2_sep_weights)|(conv5_3_sep_weights)|(conv5_4_sep_weights)|(conv5_5_sep_weights)|(conv5_6_sep_weights)|(yolo_block.0.0.0.conv.weights)|(yolo_block.0.0.1.conv.weights)|(yolo_block.0.1.0.conv.weights)|(yolo_block.0.1.1.conv.weights)|(yolo_block.1.0.0.conv.weights)|(yolo_block.1.0.1.conv.weights)|(yolo_block.1.1.0.conv.weights)|(yolo_block.1.1.1.conv.weights)|(yolo_block.1.2.conv.weights)|(yolo_block.2.0.0.conv.weights)|(yolo_block.2.0.1.conv.weights)|(yolo_block.2.1.1.conv.weights)|(yolo_block.2.2.conv.weights)|(yolo_block.2.tip.conv.weights)
```
综上，我们将MobileNetV2配置文件中的`pruned_params`设置为以下正则表达式：
```
(conv2_1_sep_weights)|(conv2_2_sep_weights)|(conv3_1_sep_weights)|(conv4_1_sep_weights)|(conv5_1_sep_weights)|(conv5_2_sep_weights)|(conv5_3_sep_weights)|(conv5_4_sep_weights)|(conv5_5_sep_weights)|(conv5_6_sep_weights)|(yolo_block.0.0.0.conv.weights)|(yolo_block.0.0.1.conv.weights)|(yolo_block.0.1.0.conv.weights)|(yolo_block.0.1.1.conv.weights)|(yolo_block.1.0.0.conv.weights)|(yolo_block.1.0.1.conv.weights)|(yolo_block.1.1.0.conv.weights)|(yolo_block.1.1.1.conv.weights)|(yolo_block.1.2.conv.weights)|(yolo_block.2.0.0.conv.weights)|(yolo_block.2.0.1.conv.weights)|(yolo_block.2.1.1.conv.weights)|(yolo_block.2.2.conv.weights)|(yolo_block.2.tip.conv.weights)
```
我们可以用上述操作观察其它检测模型的参数名称规律，然后设置合适的正则表达式来剪裁合适的参数。
## 训练
根据<a href="../../tools/train.py">PaddleDetection/tools/train.py</a>编写压缩脚本compress.py。
在该脚本中定义了Compressor对象，用于执行压缩任务。
### 执行示例
step1: 设置gpu卡
```
export CUDA_VISIBLE_DEVICES=0
```
step2: 开始训练
使用PaddleDetection提供的配置文件在用8卡进行训练：
```
python compress.py \
-s yolov3_mobilenet_v1_slim.yaml \
-c ../../configs/yolov3_mobilenet_v1_voc.yml \
-o max_iters=258 \
-d "../../dataset/voc"
```
>通过命令行覆盖设置max_iters选项，因为PaddleDetection中训练是以`batch`为单位迭代的，并没有涉及`epoch`的概念，但是PaddleSlim需要知道当前训练进行到第几个`epoch`, 所以需要将`max_iters`设置为一个`epoch`内的`batch`的数量。
如果要调整训练卡数，需要调整配置文件`yolov3_mobilenet_v1_voc.yml`中的以下参数：
- **max_iters:** 一个`epoch`中batch的数量，需要设置为`total_num / batch_size`, 其中`total_num`为训练样本总数量，`batch_size`为多卡上总的batch size.
- **YoloTrainFeed.batch_size:** 单张卡上的batch size, 受限于显存大小。
- **LeaningRate.base_lr:** 根据多卡的总`batch_size`调整`base_lr`，两者大小正相关，可以简单的按比例进行调整。
- **LearningRate.schedulers.PiecewiseDecay.milestones：**请根据batch size的变化对其调整。
- **LearningRate.schedulers.PiecewiseDecay.LinearWarmup.steps：** 请根据batch size的变化对其进行调整。
以下为4卡训练示例，通过命令行覆盖`yolov3_mobilenet_v1_voc.yml`中的参数：
```
python compress.py \
-s yolov3_mobilenet_v1_slim.yaml \
-c ../../configs/yolov3_mobilenet_v1_voc.yml \
-o max_iters=258 \
-o YoloTrainFeed.batch_size = 16 \
-d "../../dataset/voc"
```
以下为2卡训练示例，受显存所制，单卡`batch_size`不变，总`batch_size`减小，`base_lr`减小，一个epoch内batch数量增加，同时需要调整学习率相关参数，如下：
```
python compress.py \
-s yolov3_mobilenet_v1_slim.yaml \
-c ../../configs/yolov3_mobilenet_v1_voc.yml \
-o max_iters=516 \
-o LeaningRate.base_lr=0.005 \ # 0.001 /2
-o YoloTrainFeed.batch_size = 16 \
-o LearningRate.schedulers='[!PiecewiseDecay {gamma: 0.1, milestones: [110000, 124000]}, !LinearWarmup {start_factor: 0., steps: 2000}]' \
-d "../../dataset/voc"
```
通过`python compress.py --help`查看可配置参数。
通过`python ../../tools/configure.py ${option_name} help`查看如何通过命令行覆盖配置文件`yolov3_mobilenet_v1_voc.yml`中的参数。
### 保存断点（checkpoint）
如果在配置文件中设置了`checkpoint_path`, 则在压缩任务执行过程中会自动保存断点，当任务异常中断时，
重启任务会自动从`checkpoint_path`路径下按数字顺序加载最新的checkpoint文件。如果不想让重启的任务从断点恢复，
需要修改配置文件中的`checkpoint_path`，或者将`checkpoint_path`路径下文件清空。
>注意：配置文件中的信息不会保存在断点中，重启前对配置文件的修改将会生效。
## 评估
如果在配置文件中设置了`checkpoint_path`，则每个epoch会保存一个压缩后的用于评估的模型，
该模型会保存在`${checkpoint_path}/${epoch_id}/eval_model/`路径下，包含`__model__`和`__params__`两个文件。
其中，`__model__`用于保存模型结构信息，`__params__`用于保存参数（parameters）信息。
如果不需要保存评估模型，可以在定义Compressor对象时，将`save_eval_model`选项设置为False（默认为True）。
## 预测
如果在配置文件中设置了`checkpoint_path`，并且在定义Compressor对象时指定了`prune_infer_model`选项，则每个epoch都会
保存一个`inference model`。该模型是通过删除eval_program中多余的operators而得到的。
该模型会保存在`${checkpoint_path}/${epoch_id}/eval_model/`路径下，包含`__model__.infer`和`__params__`两个文件。
其中，`__model__.infer`用于保存模型结构信息，`__params__`用于保存参数（parameters）信息。
更多关于`prune_infer_model`选项的介绍，请参考：[Compressor介绍](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/docs/usage.md#121-%E5%A6%82%E4%BD%95%E6%94%B9%E5%86%99%E6%99%AE%E9%80%9A%E8%AE%AD%E7%BB%83%E8%84%9A%E6%9C%AC)
### python预测
在脚本<a href="../infer.py">PaddleDetection/tools/infer.py</a>中展示了如何使用fluid python API加载使用预测模型进行预测。
### PaddleLite
该示例中产出的预测（inference）模型可以直接用PaddleLite进行加载使用。
关于PaddleLite如何使用，请参考：[PaddleLite使用文档](https://github.com/PaddlePaddle/Paddle-Lite/wiki#%E4%BD%BF%E7%94%A8)
## 示例结果
### MobileNetV1-YOLO-V3
| FLOPS |top1_acc/top5_acc| model_size |Paddle Fluid inference time(ms)| Paddle Lite inference time(ms)|
|---|---|---|---|---|
|baseline|- |- |- |-|
|-10%|- |- |- |-|
|-30%|- |- |- |-|
|-50%|- |- |- |-|
## FAQ
slim/prune/compress.py 0 → 100644
浏览文件 @ 9a17d02c
# Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import time
import multiprocessing
import numpy as np
import sys
sys.path.append("../../")
from paddle.fluid.contrib.slim import Compressor
def set_paddle_flags(**kwargs):
for key, value in kwargs.items():
if os.environ.get(key, None) is None:
os.environ[key] = str(value)
# NOTE(paddle-dev): All of these flags should be set before
# `import paddle`. Otherwise, it would not take any effect.
set_paddle_flags(
FLAGS_eager_delete_tensor_gb=0, # enable GC to save memory
)
from paddle import fluid
from ppdet.core.workspace import load_config, merge_config, create
from ppdet.data.data_feed import create_reader
from ppdet.utils.eval_utils import parse_fetches, eval_results
from ppdet.utils.cli import ArgsParser
from ppdet.utils.check import check_gpu
import ppdet.utils.checkpoint as checkpoint
from ppdet.modeling.model_input import create_feed
import logging
FORMAT = '%(asctime)s-%(levelname)s: %(message)s'
logging.basicConfig(level=logging.INFO, format=FORMAT)
logger = logging.getLogger(__name__)
def eval_run(exe, compile_program, reader, keys, values, cls, test_feed):
"""
Run evaluation program, return program outputs.
"""
iter_id = 0
results = []
if len(cls) != 0:
values = []
for i in range(len(cls)):
_, accum_map = cls[i].get_map_var()
cls[i].reset(exe)
values.append(accum_map)
images_num = 0
start_time = time.time()
has_bbox = 'bbox' in keys
for data in reader():
data = test_feed.feed(data)
feed_data = {'image': data['image'],
'im_size': data['im_size']}
outs = exe.run(compile_program,
feed=feed_data,
fetch_list=[values[0]],
return_numpy=False)
outs.append(data['gt_box'])
outs.append(data['gt_label'])
outs.append(data['is_difficult'])
res = {
k: (np.array(v), v.recursive_sequence_lengths())
for k, v in zip(keys, outs)
}
results.append(res)
if iter_id % 100 == 0:
logger.info('Test iter {}'.format(iter_id))
iter_id += 1
images_num += len(res['bbox'][1][0]) if has_bbox else 1
logger.info('Test finish iter {}'.format(iter_id))
end_time = time.time()
fps = images_num / (end_time - start_time)
if has_bbox:
logger.info('Total number of images: {}, inference time: {} fps.'.
format(images_num, fps))
else:
logger.info('Total iteration: {}, inference time: {} batch/s.'.format(
images_num, fps))
return results
def main():
cfg = load_config(FLAGS.config)
if 'architecture' in cfg:
main_arch = cfg.architecture
else:
raise ValueError("'architecture' not specified in config file.")
merge_config(FLAGS.opt)
if 'log_iter' not in cfg:
cfg.log_iter = 20
# check if set use_gpu=True in paddlepaddle cpu version
check_gpu(cfg.use_gpu)
if cfg.use_gpu:
devices_num = fluid.core.get_cuda_device_count()
else:
devices_num = int(
os.environ.get('CPU_NUM', multiprocessing.cpu_count()))
if 'train_feed' not in cfg:
train_feed = create(main_arch + 'TrainFeed')
else:
train_feed = create(cfg.train_feed)
if 'eval_feed' not in cfg:
eval_feed = create(main_arch + 'EvalFeed')
else:
eval_feed = create(cfg.eval_feed)
place = fluid.CUDAPlace(0) if cfg.use_gpu else fluid.CPUPlace()
exe = fluid.Executor(place)
lr_builder = create('LearningRate')
optim_builder = create('OptimizerBuilder')
# build program
startup_prog = fluid.Program()
train_prog = fluid.Program()
with fluid.program_guard(train_prog, startup_prog):
with fluid.unique_name.guard():
model = create(main_arch)
train_loader, feed_vars = create_feed(train_feed, iterable=True)
train_fetches = model.train(feed_vars)
loss = train_fetches['loss']
lr = lr_builder()
optimizer = optim_builder(lr)
optimizer.minimize(loss)
train_reader = create_reader(train_feed, cfg.max_iters * devices_num,
FLAGS.dataset_dir)
train_loader.set_sample_list_generator(train_reader, place)
# parse train fetches
train_keys, train_values, _ = parse_fetches(train_fetches)
train_keys.append("lr")
train_values.append(lr.name)
train_fetch_list=[]
for k, v in zip(train_keys, train_values):
train_fetch_list.append((k, v))
eval_prog = fluid.Program()
with fluid.program_guard(eval_prog, startup_prog):
with fluid.unique_name.guard():
model = create(main_arch)
_, test_feed_vars = create_feed(eval_feed, iterable=True)
fetches = model.eval(test_feed_vars)
eval_prog = eval_prog.clone(True)
eval_reader = create_reader(eval_feed, args_path=FLAGS.dataset_dir)
#eval_pyreader.decorate_sample_list_generator(eval_reader, place)
test_data_feed = fluid.DataFeeder(test_feed_vars.values(), place)
# parse eval fetches
extra_keys = []
if cfg.metric == 'COCO':
extra_keys = ['im_info', 'im_id', 'im_shape']
if cfg.metric == 'VOC':
extra_keys = ['gt_box', 'gt_label', 'is_difficult']
eval_keys, eval_values, eval_cls = parse_fetches(fetches, eval_prog,
extra_keys)
eval_fetch_list=[]
for k, v in zip(eval_keys, eval_values):
eval_fetch_list.append((k, v))
exe.run(startup_prog)
checkpoint.load_params(exe, train_prog, cfg.pretrain_weights)
best_box_ap_list = []
def eval_func(program, scope):
#place = fluid.CPUPlace()
#exe = fluid.Executor(place)
results = eval_run(exe, program, eval_reader,
eval_keys, eval_values, eval_cls, test_data_feed)
resolution = None
if 'mask' in results[0]:
resolution = model.mask_head.resolution
box_ap_stats = eval_results(results, eval_feed, cfg.metric, cfg.num_classes,
resolution, False, FLAGS.output_eval)
if len(best_box_ap_list) == 0:
best_box_ap_list.append(box_ap_stats[0])
elif box_ap_stats[0] > best_box_ap_list[0]:
best_box_ap_list[0] = box_ap_stats[0]
checkpoint.save(exe, train_prog, os.path.join(save_dir,"best_model"))
logger.info("Best test box ap: {}".format(
best_box_ap_list[0]))
return best_box_ap_list[0]
test_feed = [('image', test_feed_vars['image'].name),
('im_size', test_feed_vars['im_size'].name)]
com = Compressor(
place,
fluid.global_scope(),
train_prog,
train_reader=train_reader,
train_feed_list=[(key, value.name) for key, value in feed_vars.items()],
train_fetch_list=train_fetch_list,
eval_program=eval_prog,
eval_reader=eval_reader,
eval_feed_list=test_feed,
eval_func={'map': eval_func},
eval_fetch_list=[eval_fetch_list[0]],
save_eval_model=True,
prune_infer_model=[["image", "im_size"],["multiclass_nms_0.tmp_0"]],
train_optimizer=None)
com.config(FLAGS.slim_file)
com.run()
if __name__ == '__main__':
parser = ArgsParser()
parser.add_argument(
"-s",
"--slim_file",
default=None,
type=str,
help="Config file of PaddleSlim.")
parser.add_argument(
"--output_eval",
default=None,
type=str,
help="Evaluation directory, default is current directory.")
parser.add_argument(
"-d",
"--dataset_dir",
default=None,
type=str,
help="Dataset path, same as DataFeed.dataset.dataset_dir")
FLAGS = parser.parse_args()
main()
slim/prune/yolov3_mobilenet_v1_slim.yaml 0 → 100644
浏览文件 @ 9a17d02c
version: 1.0
pruners:
pruner_1:
class: 'StructurePruner'
pruning_axis:
'*': 0
criterions:
'*': 'l1_norm'
strategies:
uniform_pruning_strategy:
class: 'UniformPruneStrategy'
pruner: 'pruner_1'
start_epoch: 0
target_ratio: 0.5
pruned_params: '(conv2_1_sep_weights)|(conv2_2_sep_weights)|(conv3_1_sep_weights)|(conv4_1_sep_weights)|(conv5_1_sep_weights)|(conv5_2_sep_weights)|(conv5_3_sep_weights)|(conv5_4_sep_weights)|(conv5_5_sep_weights)|(conv5_6_sep_weights)|(yolo_block.0.0.0.conv.weights)|(yolo_block.0.0.1.conv.weights)|(yolo_block.0.1.0.conv.weights)|(yolo_block.0.1.1.conv.weights)|(yolo_block.1.0.0.conv.weights)|(yolo_block.1.0.1.conv.weights)|(yolo_block.1.1.0.conv.weights)|(yolo_block.1.1.1.conv.weights)|(yolo_block.1.2.conv.weights)|(yolo_block.2.0.0.conv.weights)|(yolo_block.2.0.1.conv.weights)|(yolo_block.2.1.1.conv.weights)|(yolo_block.2.2.conv.weights)|(yolo_block.2.tip.conv.weights)'
metric_name: 'acc_top1'
compressor:
epoch: 271
eval_epoch: 10
#init_model: './checkpoints/0' # Please enable this option for loading checkpoint.
checkpoint_path: './checkpoints/'
strategies:
- uniform_pruning_strategy
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