-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathresnet.py
191 lines (147 loc) · 5.88 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
import torch.nn as nn
from utils.builder import get_builder, get_xnor_builder
from args import args
# BasicBlock {{{
class BasicBlock(nn.Module):
M = 2
expansion = 1
def __init__(self, builder, inplanes, planes, stride=1, downsample=None, base_width=64):
super(BasicBlock, self).__init__()
if base_width / 64 > 1:
raise ValueError("Base width >64 does not work for BasicBlock")
self.conv1 = builder.conv3x3(inplanes, planes, stride)
self.bn1 = builder.batchnorm(planes)
self.relu = builder.activation()
self.conv2 = builder.conv3x3(planes, planes)
self.bn2 = builder.batchnorm(planes, last_bn=True)
self.downsample = downsample
self.stride = stride
def forward(self, x):
residual = x
out = self.conv1(x)
if self.bn1 is not None:
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
if self.bn2 is not None:
out = self.bn2(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out
# BasicBlock }}}
# Bottleneck {{{
class Bottleneck(nn.Module):
M = 3
expansion = 4
def __init__(self, builder, inplanes, planes, stride=1, downsample=None, base_width=64):
super(Bottleneck, self).__init__()
width = int(planes * base_width / 64)
self.conv1 = builder.conv1x1(inplanes, width)
self.bn1 = builder.batchnorm(width)
self.conv2 = builder.conv3x3(width, width, stride=stride)
self.bn2 = builder.batchnorm(width)
self.conv3 = builder.conv1x1(width, planes * self.expansion)
self.bn3 = builder.batchnorm(planes * self.expansion, last_bn=True)
self.relu = builder.activation()
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
# Bottleneck }}}
# ResNet {{{
class ResNet(nn.Module):
def __init__(self, builder, block, layers, base_width=64):
self.inplanes = 64
super(ResNet, self).__init__()
self.base_width = base_width
self.builder = builder[0]
self.xnor_builder = builder[1]
if self.base_width // 64 > 1:
print(f"==> Using {self.base_width // 64}x wide model")
if args.first_layer_dense:
self.conv1 = nn.Conv2d(
3, self.inplanes, kernel_size=7, stride=2, padding=3, bias=False
)
else:
self.conv1 = self.builder.conv7x7(3, 64, stride=2, first_layer=True)
self.bn1 = self.builder.batchnorm(64)
self.relu = self.builder.activation()
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(self.builder, block, 64, layers[0])
self.layer2 = self._make_layer(self.builder, block, 128, layers[1], stride=2)
self.layer3 = self._make_layer(self.builder, block, 256, layers[2], stride=2)
self.layer4 = self._make_layer(self.builder, block, 512, layers[3], stride=2)
self.avgpool = nn.AdaptiveAvgPool2d(1)
if not args.instance_code:
# self.fc is the codebook
# self.fc_lr is the projection into num_bits dimensions
if args.last_layer_dense:
self.fc = nn.Conv2d(512 * block.expansion, args.num_classes, 1)
else:
if args.num_bits is None:
self.fc = self.xnor_builder.conv1x1(512 * block.expansion, args.num_classes)
else:
self.fc_lr = self.builder.conv1x1(512 * block.expansion, args.num_bits)
self.fc = self.xnor_builder.conv1x1(args.num_bits, args.num_classes)
else:
# self.fc_lr is the projection
if args.last_layer_dense:
self.fc = nn.Conv2d(512 * block.expansion, args.num_bits, 1)
else:
self.fc_lr = self.builder.conv1x1(512 * block.expansion, args.num_bits)
def _make_layer(self, builder, block, planes, blocks, stride=1):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
dconv = builder.conv1x1(
self.inplanes, planes * block.expansion, stride=stride
)
dbn = builder.batchnorm(planes * block.expansion)
if dbn is not None:
downsample = nn.Sequential(dconv, dbn)
else:
downsample = dconv
layers = []
layers.append(block(builder, self.inplanes, planes, stride, downsample, base_width=self.base_width))
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(block(builder, self.inplanes, planes, base_width=self.base_width))
return nn.Sequential(*layers)
def forward(self, x):
x = self.conv1(x)
if self.bn1 is not None:
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)
x = self.avgpool(x)
if not args.instance_code:
if args.num_bits is None:
x = self.fc(x)
else:
x = self.fc_lr(x)
x = self.fc(x)
else:
x = self.fc_lr(x)
x = x.view(x.size(0), -1)
return x
# ResNet }}}
def ResNet50(pretrained=False):
return ResNet([get_builder(), get_xnor_builder()], Bottleneck, [3, 4, 6, 3])