-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathtrident_gn.py
291 lines (229 loc) · 9.96 KB
/
trident_gn.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
from ..registry import BACKBONES
from torch.nn import init
from mmcv.cnn import constant_init, kaiming_init
from mmcv.runner import load_checkpoint
import math
import numpy as np
from .shufflenet_block import *
import logging
def conv3x3(in_planes, out_planes, stride=1):
return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
padding=1, bias=False)
class BasicBlock(nn.Module):
expansion = 1
def __init__(self, inplanes, planes, stride=1, downsample=None):
super(BasicBlock, self).__init__()
self.conv1 = conv3x3(inplanes, planes, stride)
self.bn1 = nn.GroupNorm(8, planes)
self.relu = nn.ReLU(inplace=True)
self.conv2 = conv3x3(planes, planes)
self.bn2 = nn.GroupNorm(8, planes)
self.downsample = downsample
self.stride = stride
def forward(self, x):
residual = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out
class Bottleneck(nn.Module):
expansion = 4
#CLASS torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True)
def __init__(self, inplanes, planes, stride=1, downsample=None):#inplanes输入channel,planes输出channel
super(Bottleneck, self).__init__()
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False) # change
self.bn1 = nn.GroupNorm(8, planes)
self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride, # change
padding=1, bias=False)
self.bn2 = nn.GroupNorm(8, planes)
self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)
self.bn3 = nn.GroupNorm(8, planes * 4)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
def forward(self, x):
residual = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
out = self.relu(out)
out = self.conv3(out)
out = self.bn3(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out
class trident_block(nn.Module):
expansion = 4
def __init__(self, inplanes, planes, stride=1, downsample=None, padding=[1, 2, 3], dilate=[1, 2, 3]):
super(trident_block, self).__init__()
self.stride = stride
self.padding = padding
self.dilate = dilate
self.downsample = downsample
self.share_weight4conv1 = nn.Parameter(torch.randn(planes, inplanes, 1, 1))
self.share_weight4conv2 = nn.Parameter(torch.randn(planes, planes, 3, 3))
self.share_weight4conv3 = nn.Parameter(torch.randn(planes * self.expansion, planes, 1, 1))#1*1/64, 3*3/64, 1*1/256
self.bn11 = nn.GroupNorm(8, planes)#bn层
self.bn12 = nn.GroupNorm(8, planes)
self.bn13 = nn.GroupNorm(8, planes * self.expansion)
self.bn21 = nn.GroupNorm(8, planes)
self.bn22 = nn.GroupNorm(8, planes)
self.bn23 = nn.GroupNorm(8, planes * self.expansion)
self.bn31 = nn.GroupNorm(8, planes)
self.bn32 = nn.GroupNorm(8, planes)
self.bn33 = nn.GroupNorm(8, planes * self.expansion)
self.relu1 = nn.ReLU(inplace=True)#relu层
self.relu2 = nn.ReLU(inplace=True)
self.relu3 = nn.ReLU(inplace=True)
def forward_for_small(self, x):
residual = x
out = nn.functional.conv2d(x, self.share_weight4conv1, bias=None)
out = self.bn11(out)
out = self.relu1(out)
out = nn.functional.conv2d(out, self.share_weight4conv2, bias=None, stride=self.stride, padding=self.padding[0], dilation=self.dilate[0])
out = self.bn12(out)
out = self.relu1(out)
out = nn.functional.conv2d(out, self.share_weight4conv3, bias=None)
out = self.bn13(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu1(out)
return out
def forward_for_middle(self, x):
residual = x
out = nn.functional.conv2d(x, self.share_weight4conv1, bias=None)
out = self.bn21(out)
out = self.relu2(out)
out = nn.functional.conv2d(out, self.share_weight4conv2, bias=None, stride=self.stride, padding=self.padding[1],dilation=self.dilate[1])
out = self.bn22(out)
out = self.relu2(out)
out = nn.functional.conv2d(out, self.share_weight4conv3, bias=None)
out = self.bn23(out)
if self.downsample is not None:
residual = self.downsample(x)
# print(out.shape)
# print(residual.shape)
out += residual
out = self.relu2(out)
return out
def forward_for_big(self, x):
residual = x
out = nn.functional.conv2d(x, self.share_weight4conv1, bias=None)
out = self.bn31(out)
out = self.relu3(out)
out = nn.functional.conv2d(out, self.share_weight4conv2, bias=None, stride=self.stride, padding=self.padding[2], dilation=self.dilate[2])
out = self.bn32(out)
out = self.relu3(out)
out = nn.functional.conv2d(out, self.share_weight4conv3, bias=None)#对输入平面实施2D卷积
out = self.bn33(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu3(out)
return out
def forward(self, x):
xm=x
base_feat=[]#重新定义数组
if self.downsample is not None:#衔接段需要downsample
x1 = self.forward_for_small(x)
base_feat.append(x1)
x2 = self.forward_for_middle(x)
base_feat.append(x2)
x3 = self.forward_for_big(x)
base_feat.append(x3)
else:
x1 = self.forward_for_small(xm[0])
base_feat.append(x1)
x2 = self.forward_for_middle(xm[1])
base_feat.append(x2)
x3 = self.forward_for_big(xm[2])
base_feat.append(x3)
return base_feat #三个分支
@BACKBONES.register_module
class TridentNet(nn.Module):
# def __init__(self, block, layers, num_classes=1000):#layers数组,units个数
def __init__(self, block=Bottleneck, block1=trident_block, layers=[3,4,6,3], num_classes=1000, norm_eval=True):#layers数组,units个数
self.inplanes = 64
super(TridentNet, self).__init__()
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
bias=False)
self.bn1 = nn.GroupNorm(8, 64)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=0, ceil_mode=True) # 3*3 maxpooling
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
self.layer3 = self._make_layer1(block1, 256, layers[2], stride=2)
self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
self.avgpool = nn.AvgPool2d(7)
self.fc = nn.Linear(512 * block.expansion, num_classes)#全连接分类
self.norm_eval = norm_eval
def _make_layer(self, block, planes, blocks, stride=1):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
nn.Conv2d(self.inplanes, planes * block.expansion,
kernel_size=1, stride=stride, bias=False),
nn.GroupNorm(8, planes * block.expansion),#shortcut用1*1卷积
)
layers = []
layers.append(block(self.inplanes, planes, stride, downsample))#衔接段会出现通道不匹配,需要借助downsample
self.inplanes = planes * block.expansion#维度保持一致
for i in range(1, blocks):
layers.append(block(self.inplanes, planes))#堆叠的block
return nn.Sequential(*layers)#一个resnet-unit卷积
def _make_layer1(self, block1, planes, blocks, stride=1):
downsample = None
if stride != 1 or self.inplanes != planes * block1.expansion:
downsample = nn.Sequential(
nn.Conv2d(self.inplanes, planes * block1.expansion,
kernel_size=1, stride=stride, bias=False),
nn.GroupNorm(8, planes * block1.expansion),#shortcut用1*1卷积
)
layers = []
layers.append(block1(self.inplanes, planes, stride, downsample))#衔接段会出现通道不匹配,需要借助downsample
self.inplanes = planes * block1.expansion#维度保持一致
for i in range(1, blocks):
layers.append(block1(self.inplanes, planes))#堆叠的block
return nn.Sequential(*layers)#一个trident-block卷积
def init_weights(self, pretrained=None):
if isinstance(pretrained, str):
logger = logging.getLogger()
load_checkpoint(self, pretrained, strict=False, logger=logger)
elif pretrained is None:
for m in self.modules():
if isinstance(m, nn.Conv2d):
kaiming_init(m)
elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
constant_init(m, 1)
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 = x[2]
return x
def train(self, mode=True):
super(TridentNet, self).train(mode)
if mode and self.norm_eval:
for m in self.modules():
# trick: eval have effect on BatchNorm only
if isinstance(m, (nn.BatchNorm2d)):
m.eval()