482 lines
22 KiB
Python
482 lines
22 KiB
Python
# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
|
|
#
|
|
# 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 paddle
|
|
from paddle import ParamAttr
|
|
import paddle.nn as nn
|
|
import paddle.nn.functional as F
|
|
from paddle.nn import Conv2D, BatchNorm, Linear, Dropout
|
|
from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D
|
|
from paddle.nn.initializer import Uniform
|
|
import math
|
|
|
|
__all__ = ["InceptionV3"]
|
|
|
|
|
|
class ConvBNLayer(nn.Layer):
|
|
def __init__(self,
|
|
num_channels,
|
|
num_filters,
|
|
filter_size,
|
|
stride=1,
|
|
padding=0,
|
|
groups=1,
|
|
act="relu",
|
|
name=None):
|
|
super(ConvBNLayer, self).__init__()
|
|
|
|
self.conv = Conv2D(
|
|
in_channels=num_channels,
|
|
out_channels=num_filters,
|
|
kernel_size=filter_size,
|
|
stride=stride,
|
|
padding=padding,
|
|
groups=groups,
|
|
weight_attr=ParamAttr(name=name+"_weights"),
|
|
bias_attr=False)
|
|
self.batch_norm = BatchNorm(
|
|
num_filters,
|
|
act=act,
|
|
param_attr=ParamAttr(name=name+"_bn_scale"),
|
|
bias_attr=ParamAttr(name=name+"_bn_offset"),
|
|
moving_mean_name=name+"_bn_mean",
|
|
moving_variance_name=name+"_bn_variance")
|
|
|
|
def forward(self, inputs):
|
|
y = self.conv(inputs)
|
|
y = self.batch_norm(y)
|
|
return y
|
|
|
|
class InceptionStem(nn.Layer):
|
|
def __init__(self):
|
|
super(InceptionStem, self).__init__()
|
|
self.conv_1a_3x3 = ConvBNLayer(num_channels=3,
|
|
num_filters=32,
|
|
filter_size=3,
|
|
stride=2,
|
|
act="relu",
|
|
name="conv_1a_3x3")
|
|
self.conv_2a_3x3 = ConvBNLayer(num_channels=32,
|
|
num_filters=32,
|
|
filter_size=3,
|
|
stride=1,
|
|
act="relu",
|
|
name="conv_2a_3x3")
|
|
self.conv_2b_3x3 = ConvBNLayer(num_channels=32,
|
|
num_filters=64,
|
|
filter_size=3,
|
|
padding=1,
|
|
act="relu",
|
|
name="conv_2b_3x3")
|
|
self.maxpool = MaxPool2D(kernel_size=3, stride=2, padding=0)
|
|
self.conv_3b_1x1 = ConvBNLayer(num_channels=64,
|
|
num_filters=80,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="conv_3b_1x1")
|
|
self.conv_4a_3x3 = ConvBNLayer(num_channels=80,
|
|
num_filters=192,
|
|
filter_size=3,
|
|
act="relu",
|
|
name="conv_4a_3x3")
|
|
def forward(self, x):
|
|
y = self.conv_1a_3x3(x)
|
|
y = self.conv_2a_3x3(y)
|
|
y = self.conv_2b_3x3(y)
|
|
y = self.maxpool(y)
|
|
y = self.conv_3b_1x1(y)
|
|
y = self.conv_4a_3x3(y)
|
|
y = self.maxpool(y)
|
|
return y
|
|
|
|
|
|
class InceptionA(nn.Layer):
|
|
def __init__(self, num_channels, pool_features, name=None):
|
|
super(InceptionA, self).__init__()
|
|
self.branch1x1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=64,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_a_branch1x1_"+name)
|
|
self.branch5x5_1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=48,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_a_branch5x5_1_"+name)
|
|
self.branch5x5_2 = ConvBNLayer(num_channels=48,
|
|
num_filters=64,
|
|
filter_size=5,
|
|
padding=2,
|
|
act="relu",
|
|
name="inception_a_branch5x5_2_"+name)
|
|
|
|
self.branch3x3dbl_1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=64,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_a_branch3x3dbl_1_"+name)
|
|
self.branch3x3dbl_2 = ConvBNLayer(num_channels=64,
|
|
num_filters=96,
|
|
filter_size=3,
|
|
padding=1,
|
|
act="relu",
|
|
name="inception_a_branch3x3dbl_2_"+name)
|
|
self.branch3x3dbl_3 = ConvBNLayer(num_channels=96,
|
|
num_filters=96,
|
|
filter_size=3,
|
|
padding=1,
|
|
act="relu",
|
|
name="inception_a_branch3x3dbl_3_"+name)
|
|
self.branch_pool = AvgPool2D(kernel_size=3, stride=1, padding=1, exclusive=False)
|
|
self.branch_pool_conv = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=pool_features,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_a_branch_pool_"+name)
|
|
def forward(self, x):
|
|
branch1x1 = self.branch1x1(x)
|
|
branch5x5 = self.branch5x5_1(x)
|
|
branch5x5 = self.branch5x5_2(branch5x5)
|
|
|
|
branch3x3dbl = self.branch3x3dbl_1(x)
|
|
branch3x3dbl = self.branch3x3dbl_2(branch3x3dbl)
|
|
branch3x3dbl = self.branch3x3dbl_3(branch3x3dbl)
|
|
|
|
branch_pool = self.branch_pool(x)
|
|
|
|
branch_pool = self.branch_pool_conv(branch_pool)
|
|
outputs = paddle.concat([branch1x1, branch5x5, branch3x3dbl, branch_pool], axis=1)
|
|
return outputs
|
|
|
|
|
|
class InceptionB(nn.Layer):
|
|
def __init__(self, num_channels, name=None):
|
|
super(InceptionB, self).__init__()
|
|
self.branch3x3 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=384,
|
|
filter_size=3,
|
|
stride=2,
|
|
act="relu",
|
|
name="inception_b_branch3x3_"+name)
|
|
self.branch3x3dbl_1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=64,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_b_branch3x3dbl_1_"+name)
|
|
self.branch3x3dbl_2 = ConvBNLayer(num_channels=64,
|
|
num_filters=96,
|
|
filter_size=3,
|
|
padding=1,
|
|
act="relu",
|
|
name="inception_b_branch3x3dbl_2_"+name)
|
|
self.branch3x3dbl_3 = ConvBNLayer(num_channels=96,
|
|
num_filters=96,
|
|
filter_size=3,
|
|
stride=2,
|
|
act="relu",
|
|
name="inception_b_branch3x3dbl_3_"+name)
|
|
self.branch_pool = MaxPool2D(kernel_size=3, stride=2)
|
|
|
|
def forward(self, x):
|
|
branch3x3 = self.branch3x3(x)
|
|
|
|
branch3x3dbl = self.branch3x3dbl_1(x)
|
|
branch3x3dbl = self.branch3x3dbl_2(branch3x3dbl)
|
|
branch3x3dbl = self.branch3x3dbl_3(branch3x3dbl)
|
|
|
|
branch_pool = self.branch_pool(x)
|
|
|
|
outputs = paddle.concat([branch3x3, branch3x3dbl, branch_pool], axis=1)
|
|
|
|
return outputs
|
|
|
|
class InceptionC(nn.Layer):
|
|
def __init__(self, num_channels, channels_7x7, name=None):
|
|
super(InceptionC, self).__init__()
|
|
self.branch1x1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=192,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_c_branch1x1_"+name)
|
|
self.branch7x7_1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=channels_7x7,
|
|
filter_size=1,
|
|
stride=1,
|
|
act="relu",
|
|
name="inception_c_branch7x7_1_"+name)
|
|
self.branch7x7_2 = ConvBNLayer(num_channels=channels_7x7,
|
|
num_filters=channels_7x7,
|
|
filter_size=(1, 7),
|
|
stride=1,
|
|
padding=(0, 3),
|
|
act="relu",
|
|
name="inception_c_branch7x7_2_"+name)
|
|
self.branch7x7_3 = ConvBNLayer(num_channels=channels_7x7,
|
|
num_filters=192,
|
|
filter_size=(7, 1),
|
|
stride=1,
|
|
padding=(3, 0),
|
|
act="relu",
|
|
name="inception_c_branch7x7_3_"+name)
|
|
|
|
self.branch7x7dbl_1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=channels_7x7,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_c_branch7x7dbl_1_"+name)
|
|
self.branch7x7dbl_2 = ConvBNLayer(num_channels=channels_7x7,
|
|
num_filters=channels_7x7,
|
|
filter_size=(7, 1),
|
|
padding = (3, 0),
|
|
act="relu",
|
|
name="inception_c_branch7x7dbl_2_"+name)
|
|
self.branch7x7dbl_3 = ConvBNLayer(num_channels=channels_7x7,
|
|
num_filters=channels_7x7,
|
|
filter_size=(1, 7),
|
|
padding = (0, 3),
|
|
act="relu",
|
|
name="inception_c_branch7x7dbl_3_"+name)
|
|
self.branch7x7dbl_4 = ConvBNLayer(num_channels=channels_7x7,
|
|
num_filters=channels_7x7,
|
|
filter_size=(7, 1),
|
|
padding = (3, 0),
|
|
act="relu",
|
|
name="inception_c_branch7x7dbl_4_"+name)
|
|
self.branch7x7dbl_5 = ConvBNLayer(num_channels=channels_7x7,
|
|
num_filters=192,
|
|
filter_size=(1, 7),
|
|
padding = (0, 3),
|
|
act="relu",
|
|
name="inception_c_branch7x7dbl_5_"+name)
|
|
|
|
self.branch_pool = AvgPool2D(kernel_size=3, stride=1, padding=1, exclusive=False)
|
|
self.branch_pool_conv = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=192,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_c_branch_pool_"+name)
|
|
|
|
def forward(self, x):
|
|
branch1x1 = self.branch1x1(x)
|
|
|
|
branch7x7 = self.branch7x7_1(x)
|
|
branch7x7 = self.branch7x7_2(branch7x7)
|
|
branch7x7 = self.branch7x7_3(branch7x7)
|
|
|
|
branch7x7dbl = self.branch7x7dbl_1(x)
|
|
branch7x7dbl = self.branch7x7dbl_2(branch7x7dbl)
|
|
branch7x7dbl = self.branch7x7dbl_3(branch7x7dbl)
|
|
branch7x7dbl = self.branch7x7dbl_4(branch7x7dbl)
|
|
branch7x7dbl = self.branch7x7dbl_5(branch7x7dbl)
|
|
|
|
branch_pool = self.branch_pool(x)
|
|
branch_pool = self.branch_pool_conv(branch_pool)
|
|
|
|
outputs = paddle.concat([branch1x1, branch7x7, branch7x7dbl, branch_pool], axis=1)
|
|
|
|
return outputs
|
|
|
|
class InceptionD(nn.Layer):
|
|
def __init__(self, num_channels, name=None):
|
|
super(InceptionD, self).__init__()
|
|
self.branch3x3_1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=192,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_d_branch3x3_1_"+name)
|
|
self.branch3x3_2 = ConvBNLayer(num_channels=192,
|
|
num_filters=320,
|
|
filter_size=3,
|
|
stride=2,
|
|
act="relu",
|
|
name="inception_d_branch3x3_2_"+name)
|
|
self.branch7x7x3_1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=192,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_d_branch7x7x3_1_"+name)
|
|
self.branch7x7x3_2 = ConvBNLayer(num_channels=192,
|
|
num_filters=192,
|
|
filter_size=(1, 7),
|
|
padding=(0, 3),
|
|
act="relu",
|
|
name="inception_d_branch7x7x3_2_"+name)
|
|
self.branch7x7x3_3 = ConvBNLayer(num_channels=192,
|
|
num_filters=192,
|
|
filter_size=(7, 1),
|
|
padding=(3, 0),
|
|
act="relu",
|
|
name="inception_d_branch7x7x3_3_"+name)
|
|
self.branch7x7x3_4 = ConvBNLayer(num_channels=192,
|
|
num_filters=192,
|
|
filter_size=3,
|
|
stride=2,
|
|
act="relu",
|
|
name="inception_d_branch7x7x3_4_"+name)
|
|
self.branch_pool = MaxPool2D(kernel_size=3, stride=2)
|
|
|
|
def forward(self, x):
|
|
branch3x3 = self.branch3x3_1(x)
|
|
branch3x3 = self.branch3x3_2(branch3x3)
|
|
|
|
branch7x7x3 = self.branch7x7x3_1(x)
|
|
branch7x7x3 = self.branch7x7x3_2(branch7x7x3)
|
|
branch7x7x3 = self.branch7x7x3_3(branch7x7x3)
|
|
branch7x7x3 = self.branch7x7x3_4(branch7x7x3)
|
|
|
|
branch_pool = self.branch_pool(x)
|
|
|
|
outputs = paddle.concat([branch3x3, branch7x7x3, branch_pool], axis=1)
|
|
return outputs
|
|
|
|
class InceptionE(nn.Layer):
|
|
def __init__(self, num_channels, name=None):
|
|
super(InceptionE, self).__init__()
|
|
self.branch1x1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=320,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_e_branch1x1_"+name)
|
|
self.branch3x3_1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=384,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_e_branch3x3_1_"+name)
|
|
self.branch3x3_2a = ConvBNLayer(num_channels=384,
|
|
num_filters=384,
|
|
filter_size=(1, 3),
|
|
padding=(0, 1),
|
|
act="relu",
|
|
name="inception_e_branch3x3_2a_"+name)
|
|
self.branch3x3_2b = ConvBNLayer(num_channels=384,
|
|
num_filters=384,
|
|
filter_size=(3, 1),
|
|
padding=(1, 0),
|
|
act="relu",
|
|
name="inception_e_branch3x3_2b_"+name)
|
|
|
|
self.branch3x3dbl_1 = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=448,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_e_branch3x3dbl_1_"+name)
|
|
self.branch3x3dbl_2 = ConvBNLayer(num_channels=448,
|
|
num_filters=384,
|
|
filter_size=3,
|
|
padding=1,
|
|
act="relu",
|
|
name="inception_e_branch3x3dbl_2_"+name)
|
|
self.branch3x3dbl_3a = ConvBNLayer(num_channels=384,
|
|
num_filters=384,
|
|
filter_size=(1, 3),
|
|
padding=(0, 1),
|
|
act="relu",
|
|
name="inception_e_branch3x3dbl_3a_"+name)
|
|
self.branch3x3dbl_3b = ConvBNLayer(num_channels=384,
|
|
num_filters=384,
|
|
filter_size=(3, 1),
|
|
padding=(1, 0),
|
|
act="relu",
|
|
name="inception_e_branch3x3dbl_3b_"+name)
|
|
self.branch_pool = AvgPool2D(kernel_size=3, stride=1, padding=1, exclusive=False)
|
|
self.branch_pool_conv = ConvBNLayer(num_channels=num_channels,
|
|
num_filters=192,
|
|
filter_size=1,
|
|
act="relu",
|
|
name="inception_e_branch_pool_"+name)
|
|
def forward(self, x):
|
|
branch1x1 = self.branch1x1(x)
|
|
|
|
branch3x3 = self.branch3x3_1(x)
|
|
branch3x3 = [
|
|
self.branch3x3_2a(branch3x3),
|
|
self.branch3x3_2b(branch3x3),
|
|
]
|
|
branch3x3 = paddle.concat(branch3x3, axis=1)
|
|
|
|
branch3x3dbl = self.branch3x3dbl_1(x)
|
|
branch3x3dbl = self.branch3x3dbl_2(branch3x3dbl)
|
|
branch3x3dbl = [
|
|
self.branch3x3dbl_3a(branch3x3dbl),
|
|
self.branch3x3dbl_3b(branch3x3dbl),
|
|
]
|
|
branch3x3dbl = paddle.concat(branch3x3dbl, axis=1)
|
|
|
|
branch_pool = self.branch_pool(x)
|
|
branch_pool = self.branch_pool_conv(branch_pool)
|
|
|
|
outputs = paddle.concat([branch1x1, branch3x3, branch3x3dbl, branch_pool], axis=1)
|
|
return outputs
|
|
|
|
|
|
class InceptionV3(nn.Layer):
|
|
def __init__(self, class_dim=1000):
|
|
super(InceptionV3, self).__init__()
|
|
self.inception_a_list = [[192, 256, 288], [32, 64, 64]]
|
|
self.inception_c_list = [[768, 768, 768, 768], [128, 160, 160, 192]]
|
|
|
|
self.inception_stem = InceptionStem()
|
|
self.inception_block_list = []
|
|
for i in range(len(self.inception_a_list[0])):
|
|
inception_a = self.add_sublayer("inception_a_"+str(i+1),
|
|
InceptionA(self.inception_a_list[0][i],
|
|
self.inception_a_list[1][i],
|
|
name=str(i+1)))
|
|
self.inception_block_list.append(inception_a)
|
|
inception_b = self.add_sublayer("nception_b_1",
|
|
InceptionB(288, name="1"))
|
|
self.inception_block_list.append(inception_b)
|
|
|
|
for i in range(len(self.inception_c_list[0])):
|
|
inception_c = self.add_sublayer("inception_c_"+str(i+1),
|
|
InceptionC(self.inception_c_list[0][i],
|
|
self.inception_c_list[1][i],
|
|
name=str(i+1)))
|
|
self.inception_block_list.append(inception_c)
|
|
inception_d = self.add_sublayer("inception_d_1",
|
|
InceptionD(768, name="1"))
|
|
self.inception_block_list.append(inception_d)
|
|
inception_e = self.add_sublayer("inception_e_1",
|
|
InceptionE(1280, name="1"))
|
|
self.inception_block_list.append(inception_e)
|
|
inception_e = self.add_sublayer("inception_e_2",
|
|
InceptionE(2048, name="2"))
|
|
self.inception_block_list.append(inception_e)
|
|
|
|
self.gap = AdaptiveAvgPool2D(1)
|
|
self.drop = Dropout(p=0.2, mode="downscale_in_infer")
|
|
stdv = 1.0 / math.sqrt(2048 * 1.0)
|
|
self.out = Linear(
|
|
2048,
|
|
class_dim,
|
|
weight_attr=ParamAttr(
|
|
initializer=Uniform(-stdv, stdv), name="fc_weights"),
|
|
bias_attr=ParamAttr(name="fc_offset"))
|
|
|
|
def forward(self, x):
|
|
y = self.inception_stem(x)
|
|
for inception_block in self.inception_block_list:
|
|
y = inception_block(y)
|
|
y = self.gap(y)
|
|
y = paddle.reshape(y, shape=[-1, 2048])
|
|
y = self.drop(y)
|
|
y = self.out(y)
|
|
return y
|
|
|