Skip to content

P
Paddle

BaiXuePrincess / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
未验证 提交 b38e4f28 编写于 作者: L LielinJiang 提交者： GitHub
浏览文件
操作

Refine vision models (#27476) 

* refine vision models
上级 0b4bb023
隐藏空白更改
内联 并排
Showing with 539 addition and 628 deletion
python/paddle/hapi/callbacks.py
浏览文件 @ b38e4f28
......@@ -301,10 +301,11 @@ class ProgBarLogger(Callback):
train_dataset = paddle.vision.datasets.MNIST(mode='train')
model = paddle.Model(paddle.vision.LeNet(classifier_activation=None),
lenet = paddle.vision.LeNet()
model = paddle.Model(lenet,
inputs, labels)
optim = paddle.optimizer.Adam(0.001)
optim = paddle.optimizer.Adam(0.001, parameters=lenet.parameters())
model.prepare(optimizer=optim,
loss=paddle.nn.CrossEntropyLoss(),
metrics=paddle.metric.Accuracy())
......@@ -436,10 +437,11 @@ class ModelCheckpoint(Callback):
train_dataset = paddle.vision.datasets.MNIST(mode='train')
model = paddle.Model(paddle.vision.LeNet(classifier_activation=None),
lenet = paddle.vision.LeNet()
model = paddle.Model(lenet,
inputs, labels)
optim = paddle.optimizer.Adam(0.001)
optim = paddle.optimizer.Adam(0.001, parameters=lenet.parameters())
model.prepare(optimizer=optim,
loss=paddle.nn.CrossEntropyLoss(),
metrics=paddle.metric.Accuracy())
......
python/paddle/hapi/model.py
浏览文件 @ b38e4f28
......@@ -814,10 +814,9 @@ class Model(object):
from paddle.static import InputSpec
device = paddle.set_device('cpu') # or 'gpu'
# if use static graph, do not set
paddle.disable_static(device)
net = nn.Sequential(
nn.Flatten(1),
nn.Linear(784, 200),
nn.Tanh(),
nn.Linear(200, 10))
......@@ -833,7 +832,7 @@ class Model(object):
paddle.nn.CrossEntropyLoss(),
paddle.metric.Accuracy())
data = paddle.vision.datasets.MNIST(mode='train', chw_format=False)
data = paddle.vision.datasets.MNIST(mode='train')
model.fit(data, epochs=2, batch_size=32, verbose=1)
"""
......@@ -850,7 +849,8 @@ class Model(object):
if not isinstance(inputs, (list, dict, Input)):
raise TypeError(
"'inputs' must be list or dict, and couldn't be None.")
"'inputs' must be list or dict in static graph mode")
self._inputs = self._verify_spec(inputs, True)
self._labels = self._verify_spec(labels)
......@@ -885,7 +885,6 @@ class Model(object):
from paddle.static import InputSpec
device = paddle.set_device('cpu') # or 'gpu'
paddle.disable_static(device)
net = nn.Sequential(
nn.Linear(784, 200),
......@@ -930,7 +929,6 @@ class Model(object):
from paddle.static import InputSpec
device = paddle.set_device('cpu') # or 'gpu'
paddle.disable_static(device)
net = nn.Sequential(
nn.Linear(784, 200),
......@@ -970,9 +968,12 @@ class Model(object):
import numpy as np
import paddle
import paddle.nn as nn
from paddle.static import InputSpec
device = paddle.set_device('cpu') # or 'gpu'
paddle.disable_static(device)
input = InputSpec([None, 784], 'float32', 'x')
label = InputSpec([None, 1], 'int64', 'label')
net = nn.Sequential(
nn.Linear(784, 200),
......@@ -980,7 +981,7 @@ class Model(object):
nn.Linear(200, 10),
nn.Softmax())
model = paddle.Model(net)
model = paddle.Model(net, input, label)
model.prepare()
data = np.random.random(size=(4,784)).astype(np.float32)
out = model.test_batch([data])
......@@ -1026,6 +1027,7 @@ class Model(object):
def __init__(self):
super(Mnist, self).__init__()
self.net = nn.Sequential(
nn.Flatten(1),
nn.Linear(784, 200),
nn.Tanh(),
nn.Linear(200, 10),
......@@ -1045,7 +1047,7 @@ class Model(object):
optim = paddle.optimizer.SGD(learning_rate=1e-3,
parameters=model.parameters())
model.prepare(optim, paddle.nn.CrossEntropyLoss())
data = paddle.vision.datasets.MNIST(mode='train', chw_format=False)
data = paddle.vision.datasets.MNIST(mode='train')
model.fit(data, epochs=1, batch_size=32, verbose=0)
model.save('checkpoint/test') # save for training
model.save('inference_model', False) # save for inference
......@@ -1092,15 +1094,18 @@ class Model(object):
import paddle
import paddle.nn as nn
from paddle.static import InputSpec
device = paddle.set_device('cpu')
paddle.disable_static(device)
input = InputSpec([None, 784], 'float32', 'x')
model = paddle.Model(nn.Sequential(
nn.Linear(784, 200),
nn.Tanh(),
nn.Linear(200, 10),
nn.Softmax()))
nn.Softmax()), input)
model.save('checkpoint/test')
model.load('checkpoint/test')
"""
......@@ -1165,13 +1170,15 @@ class Model(object):
import paddle
import paddle.nn as nn
from paddle.static import InputSpec
paddle.disable_static()
input = InputSpec([None, 784], 'float32', 'x')
model = paddle.Model(nn.Sequential(
nn.Linear(784, 200),
nn.Tanh(),
nn.Linear(200, 10)))
nn.Linear(200, 10)), input)
params = model.parameters()
"""
return self._adapter.parameters()
......@@ -1313,7 +1320,7 @@ class Model(object):
label = InputSpec([None, 1], 'int64', 'label')
model = paddle.Model(
paddle.vision.models.LeNet(classifier_activation=None),
paddle.vision.models.LeNet(),
input, label)
optim = paddle.optimizer.Adam(
learning_rate=0.001, parameters=model.parameters())
......@@ -1350,7 +1357,7 @@ class Model(object):
label = InputSpec([None, 1], 'int64', 'label')
model = paddle.Model(
paddle.vision.models.LeNet(classifier_activation=None), input, label)
paddle.vision.models.LeNet(), input, label)
optim = paddle.optimizer.Adam(
learning_rate=0.001, parameters=model.parameters())
model.prepare(
......@@ -1483,7 +1490,7 @@ class Model(object):
# imperative mode
paddle.disable_static()
model = paddle.Model(paddle.vision.models.LeNet())
model = paddle.Model(paddle.vision.models.LeNet(), input, label)
model.prepare(metrics=paddle.metric.Accuracy())
result = model.evaluate(val_dataset, batch_size=64)
print(result)
......@@ -1580,19 +1587,20 @@ class Model(object):
test_dataset = MnistDataset(mode='test', return_label=False)
# declarative mode
# imperative mode
input = InputSpec([-1, 1, 28, 28], 'float32', 'image')
model = paddle.Model(paddle.vision.models.LeNet(), input)
model.prepare()
result = model.predict(test_dataset, batch_size=64)
print(len(result[0]), result[0][0].shape)
# imperative mode
# declarative mode
device = paddle.set_device('cpu')
paddle.disable_static(device)
model = paddle.Model(paddle.vision.models.LeNet())
paddle.enable_static()
input = InputSpec([-1, 1, 28, 28], 'float32', 'image')
model = paddle.Model(paddle.vision.models.LeNet(), input)
model.prepare()
result = model.predict(test_dataset, batch_size=64)
print(len(result[0]), result[0][0].shape)
"""
......@@ -1832,15 +1840,11 @@ class Model(object):
import paddle
from paddle.static import InputSpec
dynamic = True
device = paddle.set_device('cpu')
paddle.disable_static(device) if dynamic else None
input = InputSpec([None, 1, 28, 28], 'float32', 'image')
label = InputSpec([None, 1], 'int64', 'label')
model = paddle.Model(paddle.vision.LeNet(classifier_activation=None),
model = paddle.Model(paddle.vision.LeNet(),
input, label)
optim = paddle.optimizer.Adam(
learning_rate=0.001, parameters=model.parameters())
......
python/paddle/metric/metrics.py
浏览文件 @ b38e4f28
......@@ -182,7 +182,6 @@ class Accuracy(Metric):
import numpy as np
import paddle
paddle.disable_static()
x = paddle.to_tensor(np.array([
[0.1, 0.2, 0.3, 0.4],
[0.1, 0.4, 0.3, 0.2],
......@@ -202,11 +201,13 @@ class Accuracy(Metric):
.. code-block:: python
import paddle
paddle.disable_static()
from paddle.static import InputSpec
input = InputSpec([None, 1, 28, 28], 'float32', 'image')
label = InputSpec([None, 1], 'int64', 'label')
train_dataset = paddle.vision.datasets.MNIST(mode='train')
model = paddle.Model(paddle.vision.LeNet(classifier_activation=None))
model = paddle.Model(paddle.vision.LeNet(), input, label)
optim = paddle.optimizer.Adam(
learning_rate=0.001, parameters=model.parameters())
model.prepare(
......
python/paddle/tests/CMakeLists.txt
浏览文件 @ b38e4f28
......@@ -8,10 +8,6 @@ foreach(TEST_OP ${DIST_TEST_OPS})
list(REMOVE_ITEM TEST_OPS ${TEST_OP})
endforeach()
# disable test_pretrained_model and test_vision_models
list(REMOVE_ITEM TEST_OPS test_pretrained_model)
list(REMOVE_ITEM TEST_OPS test_vision_models)
foreach(src ${TEST_OPS})
py_test(${src} SRCS ${src}.py)
endforeach()
......
python/paddle/tests/dist_hapi_mnist_dynamic.py
浏览文件 @ b38e4f28
......@@ -68,7 +68,7 @@ class TestDistTraning(unittest.TestCase):
inputs = [Input(im_shape, 'float32', 'image')]
labels = [Input([None, 1], 'int64', 'label')]
model = Model(LeNet(classifier_activation=None), inputs, labels)
model = Model(LeNet(), inputs, labels)
optim = fluid.optimizer.Momentum(
learning_rate=0.001, momentum=.9, parameter_list=model.parameters())
model.prepare(optim, CrossEntropyLoss(), Accuracy())
......
python/paddle/tests/dist_hapi_mnist_static.py
浏览文件 @ b38e4f28
......@@ -67,7 +67,7 @@ class TestDistTraning(unittest.TestCase):
inputs = [Input(im_shape, 'float32', 'image')]
labels = [Input([None, 1], 'int64', 'label')]
model = Model(LeNet(classifier_activation=None), inputs, labels)
model = Model(LeNet(), inputs, labels)
optim = fluid.optimizer.Momentum(
learning_rate=0.001, momentum=.9, parameter_list=model.parameters())
model.prepare(optim, CrossEntropyLoss(), Accuracy())
......
python/paddle/tests/test_model.py
浏览文件 @ b38e4f28
......@@ -40,7 +40,7 @@ from paddle.fluid.dygraph.dygraph_to_static.program_translator import ProgramTra
class LeNetDygraph(paddle.nn.Layer):
def __init__(self, num_classes=10, classifier_activation=None):
def __init__(self, num_classes=10):
super(LeNetDygraph, self).__init__()
self.num_classes = num_classes
self.features = Sequential(
......@@ -55,8 +55,7 @@ class LeNetDygraph(paddle.nn.Layer):
if num_classes > 0:
self.fc = Sequential(
Linear(400, 120), Linear(120, 84), Linear(84, 10),
Softmax()) #Todo: accept any activation
Linear(400, 120), Linear(120, 84), Linear(84, 10))
def forward(self, inputs):
x = self.features(inputs)
......@@ -67,6 +66,34 @@ class LeNetDygraph(paddle.nn.Layer):
return x
class LeNetDeclarative(fluid.dygraph.Layer):
def __init__(self, num_classes=10):
super(LeNetDeclarative, self).__init__()
self.num_classes = num_classes
self.features = Sequential(
Conv2d(
1, 6, 3, stride=1, padding=1),
ReLU(),
Pool2D(2, 'max', 2),
Conv2d(
6, 16, 5, stride=1, padding=0),
ReLU(),
Pool2D(2, 'max', 2))
if num_classes > 0:
self.fc = Sequential(
Linear(400, 120), Linear(120, 84), Linear(84, 10))
@declarative
def forward(self, inputs):
x = self.features(inputs)
if self.num_classes > 0:
x = fluid.layers.flatten(x, 1)
x = self.fc(x)
return x
class MnistDataset(MNIST):
def __init__(self, mode, return_label=True, sample_num=None):
super(MnistDataset, self).__init__(mode=mode)
......@@ -198,7 +225,7 @@ class TestModel(unittest.TestCase):
paddle.manual_seed(seed)
paddle.framework.random._manual_program_seed(seed)
net = LeNet(classifier_activation=None)
net = LeNet()
optim_new = fluid.optimizer.Adam(
learning_rate=0.001, parameter_list=net.parameters())
model = Model(net, inputs=self.inputs, labels=self.labels)
......@@ -287,14 +314,12 @@ class TestModel(unittest.TestCase):
class MyModel(paddle.nn.Layer):
def __init__(self, classifier_activation='softmax'):
def __init__(self):
super(MyModel, self).__init__()
self._fc = Linear(20, 10)
self._act = Softmax() #Todo: accept any activation
def forward(self, x):
y = self._fc(x)
y = self._act(y)
return y
......@@ -311,7 +336,7 @@ class TestModelFunction(unittest.TestCase):
def get_expect():
fluid.enable_dygraph(fluid.CPUPlace())
self.set_seed()
m = MyModel(classifier_activation=None)
m = MyModel()
optim = fluid.optimizer.SGD(learning_rate=0.001,
parameter_list=m.parameters())
m.train()
......@@ -330,7 +355,7 @@ class TestModelFunction(unittest.TestCase):
fluid.enable_dygraph(device) if dynamic else None
self.set_seed()
net = MyModel(classifier_activation=None)
net = MyModel()
optim2 = fluid.optimizer.SGD(learning_rate=0.001,
parameter_list=net.parameters())
......@@ -374,7 +399,7 @@ class TestModelFunction(unittest.TestCase):
for dynamic in [True, False]:
device = paddle.set_device('cpu')
fluid.enable_dygraph(device) if dynamic else None
net = MyModel(classifier_activation=None)
net = MyModel()
inputs = [InputSpec([None, 20], 'float32', 'x')]
labels = [InputSpec([None, 1], 'int64', 'label')]
optim = fluid.optimizer.SGD(learning_rate=0.001,
......@@ -417,7 +442,7 @@ class TestModelFunction(unittest.TestCase):
fluid.enable_dygraph(device)
inputs = [InputSpec([None, 20], 'float32', 'x')]
labels = [InputSpec([None, 1], 'int64', 'label')]
model = Model(MyModel(classifier_activation=None), inputs, labels)
model = Model(MyModel(), inputs, labels)
optim = fluid.optimizer.SGD(learning_rate=0.001,
parameter_list=model.parameters())
model.prepare(optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))
......@@ -426,7 +451,7 @@ class TestModelFunction(unittest.TestCase):
inputs = [InputSpec([None, 20], 'float32', 'x')]
labels = [InputSpec([None, 1], 'int64', 'label')]
model = Model(MyModel(classifier_activation=None), inputs, labels)
model = Model(MyModel(), inputs, labels)
optim = fluid.optimizer.SGD(learning_rate=0.001,
parameter_list=model.parameters())
model.prepare(optimizer=optim, loss=CrossEntropyLoss(reduction="sum"))
......@@ -436,7 +461,7 @@ class TestModelFunction(unittest.TestCase):
def test_static_save_dynamic_load(self):
path = tempfile.mkdtemp()
net = MyModel(classifier_activation=None)
net = MyModel()
inputs = [InputSpec([None, 20], 'float32', 'x')]
labels = [InputSpec([None, 1], 'int64', 'label')]
optim = fluid.optimizer.SGD(learning_rate=0.001,
......@@ -448,7 +473,7 @@ class TestModelFunction(unittest.TestCase):
device = paddle.set_device('cpu')
fluid.enable_dygraph(device) #if dynamic else None
net = MyModel(classifier_activation=None)
net = MyModel()
inputs = [InputSpec([None, 20], 'float32', 'x')]
labels = [InputSpec([None, 1], 'int64', 'label')]
optim = fluid.optimizer.SGD(learning_rate=0.001,
......@@ -557,7 +582,7 @@ class TestModelFunction(unittest.TestCase):
class TestRaiseError(unittest.TestCase):
def test_input_without_name(self):
net = MyModel(classifier_activation=None)
net = MyModel()
inputs = [InputSpec([None, 10], 'float32')]
labels = [InputSpec([None, 1], 'int64', 'label')]
......@@ -567,7 +592,7 @@ class TestRaiseError(unittest.TestCase):
def test_input_without_input_spec(self):
for dynamic in [True, False]:
paddle.disable_static() if dynamic else None
net = MyModel(classifier_activation=None)
net = MyModel()
with self.assertRaises(TypeError):
model = Model(net)
paddle.enable_static()
......
python/paddle/tests/test_pretrained_model.py
浏览文件 @ b38e4f28
......@@ -13,6 +13,8 @@
# limitations under the License.
import unittest
import tempfile
import shutil
import numpy as np
import paddle
......@@ -23,27 +25,36 @@ import paddle.vision.models as models
# test the predicted resutls of static graph and dynamic graph are equal
# when used pretrained model
class TestPretrainedModel(unittest.TestCase):
def infer(self, x, arch, dygraph=True):
if dygraph:
paddle.disable_static()
net = models.__dict__[arch](pretrained=True, classifier_activation=None)
inputs = [InputSpec([None, 3, 224, 224], 'float32', 'image')]
model = paddle.Model(network=net, inputs=inputs)
model.prepare()
res = model.test_batch(x)
if dygraph:
paddle.enable_static()
return res
def infer(self, arch):
path = tempfile.mkdtemp()
x = np.array(np.random.random((2, 3, 224, 224)), dtype=np.float32)
res = {}
for dygraph in [True, False]:
if not dygraph:
paddle.enable_static()
net = models.__dict__[arch]()
inputs = [InputSpec([None, 3, 224, 224], 'float32', 'image')]
model = paddle.Model(network=net, inputs=inputs)
model.prepare()
if dygraph:
model.save(path)
res['dygraph'] = model.test_batch(x)
else:
model.load(path)
res['static'] = model.test_batch(x)
if not dygraph:
paddle.disable_static()
shutil.rmtree(path)
np.testing.assert_allclose(res['dygraph'], res['static'])
def test_models(self):
arches = ['mobilenet_v1', 'mobilenet_v2', 'resnet18']
for arch in arches:
x = np.array(np.random.random((2, 3, 224, 224)), dtype=np.float32)
y_dygraph = self.infer(x, arch)
y_static = self.infer(x, arch, dygraph=False)
np.testing.assert_allclose(y_dygraph, y_static)
self.infer(arch)
if __name__ == '__main__':
......
python/paddle/tests/test_vision_models.py
浏览文件 @ b38e4f28
......@@ -36,7 +36,7 @@ class TestVisonModels(unittest.TestCase):
model.test_batch(x)
def test_mobilenetv2_pretrained(self):
self.models_infer('mobilenet_v2', pretrained=True)
self.models_infer('mobilenet_v2', pretrained=False)
def test_mobilenetv1(self):
self.models_infer('mobilenet_v1')
......
python/paddle/vision/models/lenet.py
浏览文件 @ b38e4f28
......@@ -12,20 +12,19 @@
#See the License for the specific language governing permissions and
#limitations under the License.
import paddle.fluid as fluid
from paddle.nn import Conv2d, Pool2D, Linear, ReLU, Sequential, Softmax
import paddle
import paddle.nn as nn
__all__ = ['LeNet']
class LeNet(fluid.dygraph.Layer):
class LeNet(nn.Layer):
"""LeNet model from
`"LeCun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.`_
Args:
num_classes (int): output dim of last fc layer. If num_classes <=0, last fc layer
will not be defined. Default: 10.
classifier_activation (str): activation for the last fc layer. Default: 'softmax'.
Examples:
.. code-block:: python
......@@ -35,28 +34,27 @@ class LeNet(fluid.dygraph.Layer):
model = LeNet()
"""
def __init__(self, num_classes=10, classifier_activation='softmax'):
def __init__(self, num_classes=10):
super(LeNet, self).__init__()
self.num_classes = num_classes
self.features = Sequential(
Conv2d(
self.features = nn.Sequential(
nn.Conv2d(
1, 6, 3, stride=1, padding=1),
ReLU(),
Pool2D(2, 'max', 2),
Conv2d(
nn.ReLU(),
nn.MaxPool2d(2, 2),
nn.Conv2d(
6, 16, 5, stride=1, padding=0),
ReLU(),
Pool2D(2, 'max', 2))
nn.ReLU(),
nn.MaxPool2d(2, 2))
if num_classes > 0:
self.fc = Sequential(
Linear(400, 120), Linear(120, 84), Linear(84, 10),
Softmax()) #Todo: accept any activation
self.fc = nn.Sequential(
nn.Linear(400, 120), nn.Linear(120, 84), nn.Linear(84, 10))
def forward(self, inputs):
x = self.features(inputs)
if self.num_classes > 0:
x = fluid.layers.flatten(x, 1)
x = paddle.flatten(x, 1)
x = self.fc(x)
return x
python/paddle/vision/models/mobilenetv1.py
浏览文件 @ b38e4f28
......@@ -12,10 +12,8 @@
# See the License for the specific language governing permissions and
# limitations under the License.
import paddle.fluid as fluid
from paddle.fluid.initializer import MSRA
from paddle.fluid.param_attr import ParamAttr
from paddle.fluid.dygraph.nn import Conv2D, Pool2D, BatchNorm, Linear
import paddle
import paddle.nn as nn
from paddle.utils.download import get_weights_path_from_url
......@@ -24,85 +22,66 @@ __all__ = ['MobileNetV1', 'mobilenet_v1']
model_urls = {
'mobilenetv1_1.0':
('https://paddle-hapi.bj.bcebos.com/models/mobilenet_v1_x1.0.pdparams',
'bf0d25cb0bed1114d9dac9384ce2b4a6')
'42a154c2f26f86e7457d6daded114e8c')
}
class ConvBNLayer(fluid.dygraph.Layer):
class ConvBNLayer(nn.Layer):
def __init__(self,
num_channels,
filter_size,
num_filters,
in_channels,
out_channels,
kernel_size,
stride,
padding,
channels=None,
num_groups=1,
act='relu',
use_cudnn=True,
name=None):
num_groups=1):
super(ConvBNLayer, self).__init__()
self._conv = Conv2D(
num_channels=num_channels,
num_filters=num_filters,
filter_size=filter_size,
self._conv = nn.Conv2d(
in_channels,
out_channels,
kernel_size,
stride=stride,
padding=padding,
groups=num_groups,
act=None,
use_cudnn=use_cudnn,
param_attr=ParamAttr(
initializer=MSRA(), name=self.full_name() + "_weights"),
bias_attr=False)
self._batch_norm = BatchNorm(
num_filters,
act=act,
param_attr=ParamAttr(name=self.full_name() + "_bn" + "_scale"),
bias_attr=ParamAttr(name=self.full_name() + "_bn" + "_offset"),
moving_mean_name=self.full_name() + "_bn" + '_mean',
moving_variance_name=self.full_name() + "_bn" + '_variance')
self._norm_layer = nn.BatchNorm2d(out_channels)
self._act = nn.ReLU()
def forward(self, inputs):
y = self._conv(inputs)
y = self._batch_norm(y)
return y
def forward(self, x):
x = self._conv(x)
x = self._norm_layer(x)
x = self._act(x)
return x
class DepthwiseSeparable(fluid.dygraph.Layer):
def __init__(self,
num_channels,
num_filters1,
num_filters2,
num_groups,
stride,
scale,
name=None):
class DepthwiseSeparable(nn.Layer):
def __init__(self, in_channels, out_channels1, out_channels2, num_groups,
stride, scale):
super(DepthwiseSeparable, self).__init__()
self._depthwise_conv = ConvBNLayer(
num_channels=num_channels,
num_filters=int(num_filters1 * scale),
filter_size=3,
in_channels,
int(out_channels1 * scale),
kernel_size=3,
stride=stride,
padding=1,
num_groups=int(num_groups * scale),
use_cudnn=False)
num_groups=int(num_groups * scale))
self._pointwise_conv = ConvBNLayer(
num_channels=int(num_filters1 * scale),
filter_size=1,
num_filters=int(num_filters2 * scale),
int(out_channels1 * scale),
int(out_channels2 * scale),
kernel_size=1,
stride=1,
padding=0)
def forward(self, inputs):
y = self._depthwise_conv(inputs)
y = self._pointwise_conv(y)
return y
def forward(self, x):
x = self._depthwise_conv(x)
x = self._pointwise_conv(x)
return x
class MobileNetV1(fluid.dygraph.Layer):
class MobileNetV1(nn.Layer):
"""MobileNetV1 model from
`"MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications" <https://arxiv.org/abs/1704.04861>`_.
......@@ -111,7 +90,6 @@ class MobileNetV1(fluid.dygraph.Layer):
num_classes (int): output dim of last fc layer. If num_classes <=0, last fc layer
will not be defined. Default: 1000.
with_pool (bool): use pool before the last fc layer or not. Default: True.
classifier_activation (str): activation for the last fc layer. Default: 'softmax'.
Examples:
.. code-block:: python
......@@ -121,11 +99,7 @@ class MobileNetV1(fluid.dygraph.Layer):
model = MobileNetV1()
"""
def __init__(self,
scale=1.0,
num_classes=1000,
with_pool=True,
classifier_activation='softmax'):
def __init__(self, scale=1.0, num_classes=1000, with_pool=True):
super(MobileNetV1, self).__init__()
self.scale = scale
self.dwsl = []
......@@ -133,18 +107,17 @@ class MobileNetV1(fluid.dygraph.Layer):
self.with_pool = with_pool
self.conv1 = ConvBNLayer(
num_channels=3,
filter_size=3,
channels=3,
num_filters=int(32 * scale),
in_channels=3,
out_channels=int(32 * scale),
kernel_size=3,
stride=2,
padding=1)
dws21 = self.add_sublayer(
sublayer=DepthwiseSeparable(
num_channels=int(32 * scale),
num_filters1=32,
num_filters2=64,
in_channels=int(32 * scale),
out_channels1=32,
out_channels2=64,
num_groups=32,
stride=1,
scale=scale),
......@@ -153,9 +126,9 @@ class MobileNetV1(fluid.dygraph.Layer):
dws22 = self.add_sublayer(
sublayer=DepthwiseSeparable(
num_channels=int(64 * scale),
num_filters1=64,
num_filters2=128,
in_channels=int(64 * scale),
out_channels1=64,
out_channels2=128,
num_groups=64,
stride=2,
scale=scale),
......@@ -164,9 +137,9 @@ class MobileNetV1(fluid.dygraph.Layer):
dws31 = self.add_sublayer(
sublayer=DepthwiseSeparable(
num_channels=int(128 * scale),
num_filters1=128,
num_filters2=128,
in_channels=int(128 * scale),
out_channels1=128,
out_channels2=128,
num_groups=128,
stride=1,
scale=scale),
......@@ -175,9 +148,9 @@ class MobileNetV1(fluid.dygraph.Layer):
dws32 = self.add_sublayer(
sublayer=DepthwiseSeparable(
num_channels=int(128 * scale),
num_filters1=128,
num_filters2=256,
in_channels=int(128 * scale),
out_channels1=128,
out_channels2=256,
num_groups=128,
stride=2,
scale=scale),
......@@ -186,9 +159,9 @@ class MobileNetV1(fluid.dygraph.Layer):
dws41 = self.add_sublayer(
sublayer=DepthwiseSeparable(
num_channels=int(256 * scale),
num_filters1=256,
num_filters2=256,
in_channels=int(256 * scale),
out_channels1=256,
out_channels2=256,
num_groups=256,
stride=1,
scale=scale),
......@@ -197,9 +170,9 @@ class MobileNetV1(fluid.dygraph.Layer):
dws42 = self.add_sublayer(
sublayer=DepthwiseSeparable(
num_channels=int(256 * scale),
num_filters1=256,
num_filters2=512,
in_channels=int(256 * scale),
out_channels1=256,
out_channels2=512,
num_groups=256,
stride=2,
scale=scale),
......@@ -209,9 +182,9 @@ class MobileNetV1(fluid.dygraph.Layer):
for i in range(5):
tmp = self.add_sublayer(
sublayer=DepthwiseSeparable(
num_channels=int(512 * scale),
num_filters1=512,
num_filters2=512,
in_channels=int(512 * scale),
out_channels1=512,
out_channels2=512,
num_groups=512,
stride=1,
scale=scale),
......@@ -220,9 +193,9 @@ class MobileNetV1(fluid.dygraph.Layer):
dws56 = self.add_sublayer(
sublayer=DepthwiseSeparable(
num_channels=int(512 * scale),
num_filters1=512,
num_filters2=1024,
in_channels=int(512 * scale),
out_channels1=512,
out_channels2=1024,
num_groups=512,
stride=2,
scale=scale),
......@@ -231,9 +204,9 @@ class MobileNetV1(fluid.dygraph.Layer):
dws6 = self.add_sublayer(
sublayer=DepthwiseSeparable(
num_channels=int(1024 * scale),
num_filters1=1024,
num_filters2=1024,
in_channels=int(1024 * scale),
out_channels1=1024,
out_channels2=1024,
num_groups=1024,
stride=1,
scale=scale),
......@@ -241,29 +214,23 @@ class MobileNetV1(fluid.dygraph.Layer):
self.dwsl.append(dws6)
if with_pool:
self.pool2d_avg = Pool2D(pool_type='avg', global_pooling=True)
if num_classes > -1:
self.out = Linear(
int(1024 * scale),
num_classes,
act=classifier_activation,
param_attr=ParamAttr(
initializer=MSRA(), name=self.full_name() + "fc7_weights"),
bias_attr=ParamAttr(name="fc7_offset"))
def forward(self, inputs):
y = self.conv1(inputs)
self.pool2d_avg = nn.AdaptiveAvgPool2d(1)
if num_classes > 0:
self.fc = nn.Linear(int(1024 * scale), num_classes)
def forward(self, x):
x = self.conv1(x)
for dws in self.dwsl:
y = dws(y)
x = dws(x)
if self.with_pool:
y = self.pool2d_avg(y)
x = self.pool2d_avg(x)
if self.num_classes > 0:
y = fluid.layers.reshape(y, shape=[-1, 1024])
y = self.out(y)
return y
x = paddle.flatten(x, 1)
x = self.fc(x)
return x
def _mobilenet(arch, pretrained=False, **kwargs):
......@@ -275,7 +242,7 @@ def _mobilenet(arch, pretrained=False, **kwargs):
model_urls[arch][1])
assert weight_path.endswith(
'.pdparams'), "suffix of weight must be .pdparams"
param, _ = fluid.load_dygraph(weight_path)
param, _ = paddle.load(weight_path)
model.load_dict(param)
return model
......
python/paddle/vision/models/mobilenetv2.py
浏览文件 @ b38e4f28
......@@ -14,9 +14,9 @@
import numpy as np
import paddle
import paddle.fluid as fluid
from paddle.fluid.param_attr import ParamAttr
from paddle.fluid.dygraph.nn import Conv2D, Pool2D, BatchNorm, Linear
import paddle.nn as nn
import paddle.nn.functional as F
from paddle.utils.download import get_weights_path_from_url
......@@ -25,221 +25,166 @@ __all__ = ['MobileNetV2', 'mobilenet_v2']
model_urls = {
'mobilenetv2_1.0':
('https://paddle-hapi.bj.bcebos.com/models/mobilenet_v2_x1.0.pdparams',
'8ff74f291f72533f2a7956a4efff9d88')
'0340af0a901346c8d46f4529882fb63d')
}
class ConvBNLayer(fluid.dygraph.Layer):
def __init__(self,
num_channels,
filter_size,
num_filters,
stride,
padding,
channels=None,
num_groups=1,
use_cudnn=True):
super(ConvBNLayer, self).__init__()
tmp_param = ParamAttr(name=self.full_name() + "_weights")
self._conv = Conv2D(
num_channels=num_channels,
num_filters=num_filters,
filter_size=filter_size,
stride=stride,
padding=padding,
groups=num_groups,
act=None,
use_cudnn=use_cudnn,
param_attr=tmp_param,
bias_attr=False)
self._batch_norm = BatchNorm(
num_filters,
param_attr=ParamAttr(name=self.full_name() + "_bn" + "_scale"),
bias_attr=ParamAttr(name=self.full_name() + "_bn" + "_offset"),
moving_mean_name=self.full_name() + "_bn" + '_mean',
moving_variance_name=self.full_name() + "_bn" + '_variance')
def forward(self, inputs, if_act=True):
y = self._conv(inputs)
y = self._batch_norm(y)
if if_act:
y = fluid.layers.relu6(y)
return y
class InvertedResidualUnit(fluid.dygraph.Layer):
def __init__(
self,
num_channels,
num_in_filter,
num_filters,
stride,
filter_size,
padding,
expansion_factor, ):
super(InvertedResidualUnit, self).__init__()
num_expfilter = int(round(num_in_filter * expansion_factor))
self._expand_conv = ConvBNLayer(
num_channels=num_channels,
num_filters=num_expfilter,
filter_size=1,
stride=1,
padding=0,
num_groups=1)
self._bottleneck_conv = ConvBNLayer(
num_channels=num_expfilter,
num_filters=num_expfilter,
filter_size=filter_size,
stride=stride,
padding=padding,
num_groups=num_expfilter,
use_cudnn=False)
self._linear_conv = ConvBNLayer(
num_channels=num_expfilter,
num_filters=num_filters,
filter_size=1,
stride=1,
padding=0,
num_groups=1)
def forward(self, inputs, ifshortcut):
y = self._expand_conv(inputs, if_act=True)
y = self._bottleneck_conv(y, if_act=True)
y = self._linear_conv(y, if_act=False)
if ifshortcut:
y = fluid.layers.elementwise_add(inputs, y)
return y
class InvresiBlocks(fluid.dygraph.Layer):
def __init__(self, in_c, t, c, n, s):
super(InvresiBlocks, self).__init__()
self._first_block = InvertedResidualUnit(
num_channels=in_c,
num_in_filter=in_c,
num_filters=c,
stride=s,
filter_size=3,
padding=1,
expansion_factor=t)
self._inv_blocks = []
for i in range(1, n):
tmp = self.add_sublayer(
sublayer=InvertedResidualUnit(
num_channels=c,
num_in_filter=c,
num_filters=c,
stride=1,
filter_size=3,
padding=1,
expansion_factor=t),
name=self.full_name() + "_" + str(i + 1))
self._inv_blocks.append(tmp)
def forward(self, inputs):
y = self._first_block(inputs, ifshortcut=False)
for inv_block in self._inv_blocks:
y = inv_block(y, ifshortcut=True)
return y
class MobileNetV2(fluid.dygraph.Layer):
"""MobileNetV2 model from
`"MobileNetV2: Inverted Residuals and Linear Bottlenecks" <https://arxiv.org/abs/1801.04381>`_.
Args:
scale (float): scale of channels in each layer. Default: 1.0.
num_classes (int): output dim of last fc layer. If num_classes <=0, last fc layer
will not be defined. Default: 1000.
with_pool (bool): use pool before the last fc layer or not. Default: True.
classifier_activation (str): activation for the last fc layer. Default: 'softmax'.
Examples:
.. code-block:: python
def _make_divisible(v, divisor, min_value=None):
if min_value is None:
min_value = divisor
new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
from paddle.vision.models import MobileNetV2
if new_v < 0.9 * v:
new_v += divisor
return new_v
model = MobileNetV2()
"""
class ConvBNReLU(nn.Sequential):
def __init__(self,
in_planes,
out_planes,
kernel_size=3,
stride=1,
groups=1,
norm_layer=nn.BatchNorm2d):
padding = (kernel_size - 1) // 2
super(ConvBNReLU, self).__init__(
nn.Conv2d(
in_planes,
out_planes,
kernel_size,
stride,
padding,
groups=groups,
bias_attr=False),
norm_layer(out_planes),
nn.ReLU6())
class InvertedResidual(nn.Layer):
def __init__(self,
scale=1.0,
num_classes=1000,
with_pool=True,
classifier_activation='softmax'):
inp,
oup,
stride,
expand_ratio,
norm_layer=nn.BatchNorm2d):
super(InvertedResidual, self).__init__()
self.stride = stride
assert stride in [1, 2]
hidden_dim = int(round(inp * expand_ratio))
self.use_res_connect = self.stride == 1 and inp == oup
layers = []
if expand_ratio != 1:
layers.append(
ConvBNReLU(
inp, hidden_dim, kernel_size=1, norm_layer=norm_layer))
layers.extend([
ConvBNReLU(
hidden_dim,
hidden_dim,
stride=stride,
groups=hidden_dim,
norm_layer=norm_layer),
nn.Conv2d(
hidden_dim, oup, 1, 1, 0, bias_attr=False),
norm_layer(oup),
])
self.conv = nn.Sequential(*layers)
def forward(self, x):
if self.use_res_connect:
return x + self.conv(x)
else:
return self.conv(x)
class MobileNetV2(nn.Layer):
def __init__(self, scale=1.0, num_classes=1000, with_pool=True):
"""MobileNetV2 model from
`"MobileNetV2: Inverted Residuals and Linear Bottlenecks" <https://arxiv.org/abs/1801.04381>`_.
Args:
scale (float): scale of channels in each layer. Default: 1.0.
num_classes (int): output dim of last fc layer. If num_classes <=0, last fc layer
will not be defined. Default: 1000.
with_pool (bool): use pool before the last fc layer or not. Default: True.
Examples:
.. code-block:: python
from paddle.vision.models import MobileNetV2
model = MobileNetV2()
"""
super(MobileNetV2, self).__init__()
self.scale = scale
self.num_classes = num_classes
self.with_pool = with_pool
input_channel = 32
last_channel = 1280
block = InvertedResidual
round_nearest = 8
norm_layer = nn.BatchNorm2d
inverted_residual_setting = [
[1, 16, 1, 1],
[6, 24, 2, 2],
[6, 32, 3, 2],
[6, 64, 4, 2],
[6, 96, 3, 1],
[6, 160, 3, 2],
[6, 320, 1, 1],
]
bottleneck_params_list = [
(1, 16, 1, 1),
(6, 24, 2, 2),
(6, 32, 3, 2),
(6, 64, 4, 2),
(6, 96, 3, 1),
(6, 160, 3, 2),
(6, 320, 1, 1),
input_channel = _make_divisible(input_channel * scale, round_nearest)
self.last_channel = _make_divisible(last_channel * max(1.0, scale),
round_nearest)
features = [
ConvBNReLU(
3, input_channel, stride=2, norm_layer=norm_layer)
]
self._conv1 = ConvBNLayer(
num_channels=3,
num_filters=int(32 * scale),
filter_size=3,
stride=2,
padding=1)
self._invl = []
i = 1
in_c = int(32 * scale)
for layer_setting in bottleneck_params_list:
t, c, n, s = layer_setting
i += 1
tmp = self.add_sublayer(
sublayer=InvresiBlocks(
in_c=in_c, t=t, c=int(c * scale), n=n, s=s),
name='conv' + str(i))
self._invl.append(tmp)
in_c = int(c * scale)
self._out_c = int(1280 * scale) if scale > 1.0 else 1280
self._conv9 = ConvBNLayer(
num_channels=in_c,
num_filters=self._out_c,
filter_size=1,
stride=1,
padding=0)
for t, c, n, s in inverted_residual_setting:
output_channel = _make_divisible(c * scale, round_nearest)
for i in range(n):
stride = s if i == 0 else 1
features.append(
block(
input_channel,
output_channel,
stride,
expand_ratio=t,
norm_layer=norm_layer))
input_channel = output_channel
features.append(
ConvBNReLU(
input_channel,
self.last_channel,
kernel_size=1,
norm_layer=norm_layer))
self.features = nn.Sequential(*features)
if with_pool:
self._pool2d_avg = Pool2D(pool_type='avg', global_pooling=True)
if num_classes > 0:
tmp_param = ParamAttr(name=self.full_name() + "fc10_weights")
self._fc = Linear(
self._out_c,
num_classes,
act=classifier_activation,
param_attr=tmp_param,
bias_attr=ParamAttr(name="fc10_offset"))
def forward(self, inputs):
y = self._conv1(inputs, if_act=True)
for inv in self._invl:
y = inv(y)
y = self._conv9(y, if_act=True)
self.pool2d_avg = nn.AdaptiveAvgPool2d(1)
if self.num_classes > 0:
self.classifier = nn.Sequential(
nn.Dropout(0.2), nn.Linear(self.last_channel, num_classes))
def forward(self, x):
x = self.features(x)
if self.with_pool:
y = self._pool2d_avg(y)
x = self.pool2d_avg(x)
if self.num_classes > 0:
y = fluid.layers.reshape(y, shape=[-1, self._out_c])
y = self._fc(y)
return y
x = paddle.flatten(x, 1)
x = self.classifier(x)
return x
def _mobilenet(arch, pretrained=False, **kwargs):
......@@ -251,7 +196,7 @@ def _mobilenet(arch, pretrained=False, **kwargs):
model_urls[arch][1])
assert weight_path.endswith(
'.pdparams'), "suffix of weight must be .pdparams"
param, _ = fluid.load_dygraph(weight_path)
param, _ = paddle.load(weight_path)
model.load_dict(param)
return model
......
python/paddle/vision/models/resnet.py
浏览文件 @ b38e4f28
......@@ -15,11 +15,8 @@
from __future__ import division
from __future__ import print_function
import math
import paddle.fluid as fluid
from paddle.fluid.dygraph.nn import Conv2D, Pool2D, BatchNorm, Linear
from paddle.fluid.dygraph.container import Sequential
import paddle
import paddle.nn as nn
from paddle.utils.download import get_weights_path_from_url
......@@ -29,143 +26,129 @@ __all__ = [
model_urls = {
'resnet18': ('https://paddle-hapi.bj.bcebos.com/models/resnet18.pdparams',
'0ba53eea9bc970962d0ef96f7b94057e'),
'cf548f46534aa3560945be4b95cd11c4'),
'resnet34': ('https://paddle-hapi.bj.bcebos.com/models/resnet34.pdparams',
'46bc9f7c3dd2e55b7866285bee91eff3'),
'8d2275cf8706028345f78ac0e1d31969'),
'resnet50': ('https://paddle-hapi.bj.bcebos.com/models/resnet50.pdparams',
'5ce890a9ad386df17cf7fe2313dca0a1'),
'ca6f485ee1ab0492d38f323885b0ad80'),
'resnet101': ('https://paddle-hapi.bj.bcebos.com/models/resnet101.pdparams',
'fb07a451df331e4b0bb861ed97c3a9b9'),
'02f35f034ca3858e1e54d4036443c92d'),
'resnet152': ('https://paddle-hapi.bj.bcebos.com/models/resnet152.pdparams',
'f9c700f26d3644bb76ad2226ed5f5713'),
'7ad16a2f1e7333859ff986138630fd7a'),
}
class ConvBNLayer(fluid.dygraph.Layer):
class BasicBlock(nn.Layer):
expansion = 1
def __init__(self,
num_channels,
num_filters,
filter_size,
inplanes,
planes,
stride=1,
downsample=None,
groups=1,
act=None):
super(ConvBNLayer, self).__init__()
self._conv = Conv2D(
num_channels=num_channels,
num_filters=num_filters,
filter_size=filter_size,
stride=stride,
padding=(filter_size - 1) // 2,
groups=groups,
act=None,
bias_attr=False)
self._batch_norm = BatchNorm(num_filters, act=act)
def forward(self, inputs):
x = self._conv(inputs)
x = self._batch_norm(x)
return x
class BasicBlock(fluid.dygraph.Layer):
"""residual block of resnet18 and resnet34
"""
expansion = 1
def __init__(self, num_channels, num_filters, stride, shortcut=True):
base_width=64,
dilation=1,
norm_layer=None):
super(BasicBlock, self).__init__()
if norm_layer is None:
norm_layer = nn.BatchNorm2d
self.conv0 = ConvBNLayer(
num_channels=num_channels,
num_filters=num_filters,
filter_size=3,
act='relu')
self.conv1 = ConvBNLayer(
num_channels=num_filters,
num_filters=num_filters,
filter_size=3,
stride=stride,
act='relu')
if dilation > 1:
raise NotImplementedError(
"Dilation > 1 not supported in BasicBlock")
if not shortcut:
self.short = ConvBNLayer(
num_channels=num_channels,
num_filters=num_filters,
filter_size=1,
stride=stride)
self.conv1 = nn.Conv2d(
inplanes, planes, 3, padding=1, stride=stride, bias_attr=False)
self.bn1 = norm_layer(planes)
self.relu = nn.ReLU()
self.conv2 = nn.Conv2d(planes, planes, 3, padding=1, bias_attr=False)
self.bn2 = norm_layer(planes)
self.downsample = downsample
self.stride = stride
self.shortcut = shortcut
def forward(self, x):
identity = x
def forward(self, inputs):
y = self.conv0(inputs)
conv1 = self.conv1(y)
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
if self.shortcut:
short = inputs
else:
short = self.short(inputs)
out = self.conv2(out)
out = self.bn2(out)
y = short + conv1
if self.downsample is not None:
identity = self.downsample(x)
return fluid.layers.relu(y)
out += identity
out = self.relu(out)
return out
class BottleneckBlock(fluid.dygraph.Layer):
"""residual block of resnet50, resnet101 amd resnet152
"""
class BottleneckBlock(nn.Layer):
expansion = 4
def __init__(self, num_channels, num_filters, stride, shortcut=True):
def __init__(self,
inplanes,
planes,
stride=1,
downsample=None,
groups=1,
base_width=64,
dilation=1,
norm_layer=None):
super(BottleneckBlock, self).__init__()
self.conv0 = ConvBNLayer(
num_channels=num_channels,
num_filters=num_filters,
filter_size=1,
act='relu')
self.conv1 = ConvBNLayer(
num_channels=num_filters,
num_filters=num_filters,
filter_size=3,
if norm_layer is None:
norm_layer = nn.BatchNorm2d
width = int(planes * (base_width / 64.)) * groups
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,
act='relu')
self.conv2 = ConvBNLayer(
num_channels=num_filters,
num_filters=num_filters * self.expansion,
filter_size=1,
act=None)
groups=groups,
dilation=dilation,
bias_attr=False)
self.bn2 = norm_layer(width)
if not shortcut:
self.short = ConvBNLayer(
num_channels=num_channels,
num_filters=num_filters * self.expansion,
filter_size=1,
stride=stride)
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
self.stride = stride
self.shortcut = shortcut
def forward(self, x):
identity = x
self._num_channels_out = num_filters * self.expansion
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
def forward(self, inputs):
x = self.conv0(inputs)
conv1 = self.conv1(x)
conv2 = self.conv2(conv1)
out = self.conv2(out)
out = self.bn2(out)
out = self.relu(out)
if self.shortcut:
short = inputs
else:
short = self.short(inputs)
out = self.conv3(out)
out = self.bn3(out)
x = fluid.layers.elementwise_add(x=short, y=conv2)
if self.downsample is not None:
identity = self.downsample(x)
return fluid.layers.relu(x)
out += identity
out = self.relu(out)
return out
class ResNet(fluid.dygraph.Layer):
class ResNet(nn.Layer):
"""ResNet model from
`"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_
......@@ -175,7 +158,6 @@ class ResNet(fluid.dygraph.Layer):
num_classes (int): output dim of last fc layer. If num_classes <=0, last fc layer
will not be defined. Default: 1000.
with_pool (bool): use pool before the last fc layer or not. Default: True.
classifier_activation (str): activation for the last fc layer. Default: 'softmax'.
Examples:
.. code-block:: python
......@@ -189,82 +171,87 @@ class ResNet(fluid.dygraph.Layer):
"""
def __init__(self,
Block,
depth=50,
num_classes=1000,
with_pool=True,
classifier_activation='softmax'):
def __init__(self, block, depth, num_classes=1000, with_pool=True):
super(ResNet, self).__init__()
self.num_classes = num_classes
self.with_pool = with_pool
layer_config = {
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],
152: [3, 8, 36, 3]
}
assert depth in layer_config.keys(), \
"supported depth are {} but input layer is {}".format(
layer_config.keys(), depth)
layers = layer_config[depth]
in_channels = 64
out_channels = [64, 128, 256, 512]
self.conv = ConvBNLayer(
num_channels=3, num_filters=64, filter_size=7, stride=2, act='relu')
self.pool = Pool2D(
pool_size=3, pool_stride=2, pool_padding=1, pool_type='max')
self.layers = []
for idx, num_blocks in enumerate(layers):
blocks = []
shortcut = False
for b in range(num_blocks):
if b == 1:
in_channels = out_channels[idx] * Block.expansion
block = Block(
num_channels=in_channels,
num_filters=out_channels[idx],
stride=2 if b == 0 and idx != 0 else 1,
shortcut=shortcut)
blocks.append(block)
shortcut = True
layer = self.add_sublayer("layer_{}".format(idx),
Sequential(*blocks))
self.layers.append(layer)
layers = layer_cfg[depth]
self.num_classes = num_classes
self.with_pool = with_pool
self._norm_layer = nn.BatchNorm2d
self.inplanes = 64
self.dilation = 1
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)
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
if with_pool:
self.global_pool = Pool2D(
pool_size=7, pool_type='avg', global_pooling=True)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
if num_classes > 0:
stdv = 1.0 / math.sqrt(out_channels[-1] * Block.expansion * 1.0)
self.fc_input_dim = out_channels[-1] * Block.expansion * 1 * 1
self.fc = Linear(
self.fc_input_dim,
num_classes,
act=classifier_activation,
param_attr=fluid.param_attr.ParamAttr(
initializer=fluid.initializer.Uniform(-stdv, stdv)))
def forward(self, inputs):
x = self.conv(inputs)
x = self.pool(x)
for layer in self.layers:
x = layer(x)
if self.with_pool:
x = self.global_pool(x)
if self.num_classes > -1:
x = fluid.layers.reshape(x, shape=[-1, self.fc_input_dim])
self.fc = nn.Linear(512 * block.expansion, num_classes)
def _make_layer(self, block, planes, blocks, stride=1, dilate=False):
norm_layer = self._norm_layer
downsample = None
previous_dilation = self.dilation
if dilate:
self.dilation *= stride
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),
norm_layer(planes * block.expansion), )
layers = []
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))
return nn.Sequential(*layers)
def forward(self, x):
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
if self.with_pool > 0:
x = self.avgpool(x)
if self.num_classes > 0:
x = paddle.flatten(x, 1)
x = self.fc(x)
return x
......@@ -277,7 +264,7 @@ def _resnet(arch, Block, depth, pretrained, **kwargs):
model_urls[arch][1])
assert weight_path.endswith(
'.pdparams'), "suffix of weight must be .pdparams"
param, _ = fluid.load_dygraph(weight_path)
param, _ = paddle.load(weight_path)
model.set_dict(param)
return model
......
python/paddle/vision/models/vgg.py
浏览文件 @ b38e4f28
......@@ -12,9 +12,8 @@
# See the License for the specific language governing permissions and
# limitations under the License.
import paddle.fluid as fluid
from paddle.nn import Conv2d, Pool2D, BatchNorm, Linear, ReLU, Softmax
from paddle.fluid.dygraph.container import Sequential
import paddle
import paddle.nn as nn
from paddle.utils.download import get_weights_path_from_url
......@@ -28,39 +27,18 @@ __all__ = [
model_urls = {
'vgg16': ('https://paddle-hapi.bj.bcebos.com/models/vgg16.pdparams',
'c788f453a3b999063e8da043456281ee')
'89bbffc0f87d260be9b8cdc169c991c4')
}
class Classifier(fluid.dygraph.Layer):
def __init__(self, num_classes, classifier_activation='softmax'):
super(Classifier, self).__init__()
self.linear1 = Linear(512 * 7 * 7, 4096)
self.linear2 = Linear(4096, 4096)
self.linear3 = Linear(4096, num_classes)
self.act = Softmax() #Todo: accept any activation
def forward(self, x):
x = self.linear1(x)
x = fluid.layers.relu(x)
x = fluid.layers.dropout(x, 0.5)
x = self.linear2(x)
x = fluid.layers.relu(x)
x = fluid.layers.dropout(x, 0.5)
x = self.linear3(x)
out = self.act(x)
return out
class VGG(fluid.dygraph.Layer):
class VGG(nn.Layer):
"""VGG model from
`"Very Deep Convolutional Networks For Large-Scale Image Recognition" <https://arxiv.org/pdf/1409.1556.pdf>`_
Args:
features (fluid.dygraph.Layer): vgg features create by function make_layers.
features (nn.Layer): vgg features create by function make_layers.
num_classes (int): output dim of last fc layer. If num_classes <=0, last fc layer
will not be defined. Default: 1000.
classifier_activation (str): activation for the last fc layer. Default: 'softmax'.
Examples:
.. code-block:: python
......@@ -76,44 +54,41 @@ class VGG(fluid.dygraph.Layer):
"""
def __init__(self,
features,
num_classes=1000,
classifier_activation='softmax'):
def __init__(self, features, num_classes=1000):
super(VGG, self).__init__()
self.features = features
self.num_classes = num_classes
if num_classes > 0:
classifier = Classifier(num_classes, classifier_activation)
self.classifier = self.add_sublayer("classifier",
Sequential(classifier))
self.avgpool = nn.AdaptiveAvgPool2d((7, 7))
self.classifier = nn.Sequential(
nn.Linear(512 * 7 * 7, 4096),
nn.ReLU(),
nn.Dropout(),
nn.Linear(4096, 4096),
nn.ReLU(),
nn.Dropout(),
nn.Linear(4096, num_classes), )
def forward(self, x):
x = self.features(x)
if self.num_classes > 0:
x = fluid.layers.flatten(x, 1)
x = self.classifier(x)
x = self.avgpool(x)
x = paddle.flatten(x, 1)
x = self.classifier(x)
return x
def make_layers(cfg, batch_norm=False):
layers = []
in_channels = 3
for v in cfg:
if v == 'M':
layers += [Pool2D(pool_size=2, pool_stride=2)]
layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
else:
conv2d = nn.Conv2d(in_channels, v, kernel_size=3, padding=1)
if batch_norm:
conv2d = Conv2d(in_channels, v, kernel_size=3, padding=1)
layers += [conv2d, BatchNorm(v), ReLU()]
layers += [conv2d, nn.BatchNorm2d(v), nn.ReLU()]
else:
conv2d = Conv2d(in_channels, v, kernel_size=3, padding=1)
layers += [conv2d, ReLU()]
layers += [conv2d, nn.ReLU()]
in_channels = v
return Sequential(*layers)
return nn.Sequential(*layers)
cfgs = {
......@@ -144,7 +119,7 @@ def _vgg(arch, cfg, batch_norm, pretrained, **kwargs):
model_urls[arch][1])
assert weight_path.endswith(
'.pdparams'), "suffix of weight must be .pdparams"
param, _ = fluid.load_dygraph(weight_path)
param, _ = paddle.load(weight_path)
model.load_dict(param)
return model
......
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