-
Notifications
You must be signed in to change notification settings - Fork 2
/
resnet.py
239 lines (185 loc) · 7.89 KB
/
resnet.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
from functools import partial
from timm.models import ResNet
from timm.models.resnet import Bottleneck, BasicBlock, _create_resnet, default_cfgs
import math
import copy
from timm.models.layers import DropBlock2d, DropPath, AvgPool2dSame, BlurPool2d, create_attn, create_classifier
from timm.models.registry import register_model
import torch
import torch.nn as nn
from timm.models.helpers import build_model_with_cfg
from utils import my_scaler
class GREBottleneck(Bottleneck):
def __init__(self, inplanes, planes, stride=1, downsample=None, cardinality=1, base_width=64,
reduce_first=1, dilation=1, first_dilation=None, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d,
attn_layer=None, aa_layer=None, drop_block=None, drop_path=None):
super(GREBottleneck, self).__init__(inplanes, planes, stride, downsample, cardinality, base_width,
reduce_first, dilation, first_dilation, act_layer, norm_layer,
attn_layer, aa_layer, drop_block, drop_path)
def zero_init_last_bn(self):
nn.init.zeros_(self.bn3.weight)
def forward(self, x):
scale = my_scaler.get_scale()
residual = x
x = self.conv1(x)
x = self.bn1(x)
if self.drop_block is not None:
x = self.drop_block(x)
x = self.act1(x)
x = self.conv2(x)
x = self.bn2(x)
if self.drop_block is not None:
x = self.drop_block(x)
x = self.act2(x)
if self.aa is not None:
x = self.aa(x)
x = self.conv3(x)
x = self.bn3(x)
if self.drop_block is not None:
x = self.drop_block(x)
if self.se is not None:
x = self.se(x)
# if self.drop_path is not None:
# x = self.drop_path(x)
if self.downsample is not None:
residual = self.downsample(residual)
x = scale * residual + x
x = self.act3(x)
return x
class GREBasicBlock(BasicBlock):
expansion = 1
def __init__(self, inplanes, planes, stride=1, downsample=None, cardinality=1, base_width=64,
reduce_first=1, dilation=1, first_dilation=None, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d,
attn_layer=None, aa_layer=None, drop_block=None, drop_path=None):
super(GREBasicBlock, self).__init__(inplanes, planes, stride, downsample, cardinality, base_width, reduce_first,
dilation, first_dilation, act_layer, norm_layer, attn_layer, aa_layer,
drop_block, drop_path)
def forward(self, x):
if self.downsample is not None:
residual = x
x = self.conv1(x)
x = self.bn1(x)
if self.drop_block is not None:
x = self.drop_block(x)
x = self.act1(x)
if self.aa is not None:
x = self.aa(x)
x = self.conv2(x)
x = self.bn2(x)
if self.drop_block is not None:
x = self.drop_block(x)
if self.se is not None:
x = self.se(x)
if self.drop_path is not None:
x = self.drop_path(x)
if self.downsample is not None:
residual = self.downsample(residual)
x += residual
x = self.act2(x)
return x
else:
scale = my_scaler.get_scale()
residual = x
x = self.conv1(x)
x = self.bn1(x)
if self.drop_block is not None:
x = self.drop_block(x)
x = scale * residual + x
x = self.act1(x)
if self.aa is not None:
x = self.aa(x)
residual = x
x = self.conv2(x)
x = self.bn2(x)
if self.drop_block is not None:
x = self.drop_block(x)
if self.se is not None:
x = self.se(x)
if self.drop_path is not None:
x = self.drop_path(x)
x += scale * residual
x = self.act2(x)
return x
class GREBasicBlockV2(BasicBlock):
expansion = 1
def __init__(self, inplanes, planes, stride=1, downsample=None, cardinality=1, base_width=64,
reduce_first=1, dilation=1, first_dilation=None, act_layer=nn.ReLU, norm_layer=nn.BatchNorm2d,
attn_layer=None, aa_layer=None, drop_block=None, drop_path=None):
super(GREBasicBlockV2, self).__init__(inplanes, planes, stride, downsample, cardinality, base_width, reduce_first,
dilation, first_dilation, act_layer, norm_layer, attn_layer, aa_layer,
drop_block, drop_path)
if self.downsample is None:
first_planes = planes // reduce_first
outplanes = planes * self.expansion
self.conv_11 = nn.ModuleList()
self.conv_11.append(nn.Sequential(nn.Conv2d(inplanes, first_planes, 1, bias=False), nn.BatchNorm2d(first_planes)))
self.conv_11.append(nn.Sequential(nn.Conv2d(first_planes, outplanes, 1, bias=False), nn.BatchNorm2d(outplanes)))
self.identity = nn.ModuleList()
self.identity.append(nn.BatchNorm2d(first_planes))
self.identity.append(nn.BatchNorm2d(outplanes))
def forward(self, x):
if self.downsample is not None:
residual = x
x = self.conv1(x)
x = self.bn1(x)
if self.drop_block is not None:
x = self.drop_block(x)
x = self.act1(x)
if self.aa is not None:
x = self.aa(x)
x = self.conv2(x)
x = self.bn2(x)
if self.drop_block is not None:
x = self.drop_block(x)
if self.se is not None:
x = self.se(x)
if self.drop_path is not None:
x = self.drop_path(x)
if self.downsample is not None:
residual = self.downsample(residual)
x += residual
x = self.act2(x)
return x
else:
scale = my_scaler.get_scale()
residual = x
x = self.conv1(x)
x = self.bn1(x)
if self.drop_block is not None:
x = self.drop_block(x)
x = scale * (self.identity[0](residual) + self.conv_11[0](residual)) + x
x = self.act1(x)
if self.aa is not None:
x = self.aa(x)
residual = x
x = self.conv2(x)
x = self.bn2(x)
if self.drop_block is not None:
x = self.drop_block(x)
if self.se is not None:
x = self.se(x)
if self.drop_path is not None:
x = self.drop_path(x)
x = scale * (self.identity[1](residual) + self.conv_11[1](residual)) + x
x = self.act2(x)
return x
def _create_gre_resnet(variant, pretrained=False, **kwargs):
return build_model_with_cfg(ResNet, variant, default_cfg=default_cfgs[variant], pretrained=pretrained, **kwargs)
@register_model
def gre_resnet50(pretrained=False, **kwargs):
"""Constructs a GRE ResNet-50 model.
"""
model_args = dict(block=GREBottleneck, layers=[3, 4, 6, 3], **kwargs)
return _create_gre_resnet('resnet50', pretrained, **model_args)
@register_model
def gre_resnet34(pretrained=False, **kwargs):
"""Constructs a ResNet-34 model.
"""
model_args = dict(block=GREBasicBlock, layers=[3, 4, 6, 3], **kwargs)
return _create_gre_resnet('resnet34', pretrained, **model_args)
@register_model
def gre_resnet34v2(pretrained=False, **kwargs):
"""Constructs a ResNet-34 model.
"""
model_args = dict(block=GREBasicBlockV2, layers=[3, 4, 6, 3], **kwargs)
return _create_gre_resnet('resnet34', pretrained, **model_args)