forked from chenxuluo/GST-video
-
Notifications
You must be signed in to change notification settings - Fork 0
/
GST.py
179 lines (145 loc) · 6.09 KB
/
GST.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
import torch
import torch.nn as nn
import math
import torch.utils.model_zoo as model_zoo
import torchvision.models as models
__all__ = ['ResNet', 'resnet50', 'resnet101','resnet152']
model_urls = {
'resnet18': 'https://download.pytorch.org/models/resnet18-5c106cde.pth',
'resnet34': 'https://download.pytorch.org/models/resnet34-333f7ec4.pth',
'resnet50': 'https://download.pytorch.org/models/resnet50-19c8e357.pth',
'resnet101': 'https://download.pytorch.org/models/resnet101-5d3b4d8f.pth',
'resnet152': 'https://download.pytorch.org/models/resnet152-b121ed2d.pth',
}
class Bottleneck(nn.Module):
expansion = 4
def __init__(self, inplanes, planes, alpha, beta, stride = 1, downsample = None):
super(Bottleneck, self).__init__()
self.conv1 = nn.Conv3d(inplanes, planes, kernel_size=1, bias=False) #定义一个逐点卷积进行通道融合
self.bn1 = nn.BatchNorm3d(planes) #定义一个3DBN操作
#定义2D的空间卷积Kernal_size = 1X3X3
self.conv2 = nn.Conv3d(planes// beta, planes//alpha*(alpha-1), kernel_size=(1,3,3), stride=(1,stride,stride),
padding=(0,1,1), bias=False)
#定义时间卷积Kernal_size = 3X3X3
self.Tconv = nn.Conv3d(planes//beta, planes//alpha, kernel_size = 3, bias = False,stride=(1,stride,stride), #时域使用的3D卷积并未使用
padding = (1,1,1))
self.bn2 = nn.BatchNorm3d(planes)#通道数没变！经过GST后通道数还是planes
self.conv3 = nn.Conv3d(planes, planes * self.expansion, kernel_size=1, bias=False) #再一次进行3D的逐点卷积扩充通道
self.bn3 = nn.BatchNorm3d(planes * self.expansion)#在一次进行3DBN
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
self.alpha = alpha
self.beta = beta
def forward(self, x):
residual = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out) #out ==> [batch_size,C,T,W,H]
if self.beta == 2:
nchannels = out.size()[1] // self.beta #out.size()[1]即为输出的通道数
left = out[:,:nchannels] #将通道进行拆分
right = out[:,nchannels:]
out1 = self.conv2(left)
out2 = self.Tconv(right)
else:
out1 = self.conv2(out)
out2 = self.Tconv(out)
out = torch.cat((out1,out2),dim=1) #在通道维度上进行拼接索引1即为通道
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(residual)
out += residual #加上残差
out = self.relu(out) #再进行激活一次
return out
class ResNet(nn.Module):
def __init__(self, block, layers, alpha, beta, num_classes=1000):
self.inplanes = 64
super(ResNet, self).__init__()
self.conv1 = nn.Conv3d(3, 64, kernel_size=(1,7,7), stride=(1,2,2), padding=(0,3,3),
bias=False)
self.bn1 = nn.BatchNorm3d(64)
self.relu = nn.ReLU(inplace=True) #定义Relu激活
self.maxpool = nn.MaxPool3d(kernel_size=(1,3,3), stride=(1,2,2), padding=(0,1,1))
self.layer1 = self._make_layer(block, 64, layers[0], alpha, beta)
self.layer2 = self._make_layer(block, 128, layers[1], alpha, beta, stride=2)
self.layer3 = self._make_layer(block, 256, layers[2], alpha, beta, stride=2)
self.layer4 = self._make_layer(block, 512, layers[3], alpha, beta, stride=2)
self.avgpool = nn.AvgPool2d(7, stride=1)
self.fc = nn.Linear(512 * block.expansion, num_classes)
for m in self.modules():
if isinstance(m, nn.Conv3d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
elif isinstance(m, nn.BatchNorm3d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
def _make_layer(self, block, planes, blocks, alpha, beta, stride=1):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
nn.Conv3d(self.inplanes, planes * block.expansion,
kernel_size=1, stride=(1,stride,stride), bias=False),
nn.BatchNorm3d(planes * block.expansion),
)
layers = []
layers.append(block(self.inplanes, planes, alpha, beta, stride, downsample))
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(block(self.inplanes, planes, alpha, beta))
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)
print("layer4_out.size()",x.size())
x = x.transpose(1,2).contiguous()
x = x.view((-1,)+x.size()[2:]) #resize成（N*T,C,W,H）
print("x.view()",x.size())
x = self.avgpool(x)
print("avg:",x.size())
x = x.view(x.size(0), -1)
x = self.fc(x)
print("out.size():",x.size())
return x
def resnet50(alpha, beta,**kwargs):
"""Constructs a ResNet-50 based model.
"""
model = ResNet(Bottleneck, [3, 4, 6, 3], alpha, beta, **kwargs)
checkpoint = model_zoo.load_url(model_urls['resnet50'],model_dir='./pretrained/') #加载Imagenet预训练的模型参数
layer_name = list(checkpoint.keys()) #checkpoint是一个字典{'layer_name','权重tensor'}
for ln in layer_name:
if 'conv' in ln or 'downsample.0.weight' in ln:
checkpoint[ln] = checkpoint[ln].unsqueeze(2) #扩充维度，很重要！
if 'conv2' in ln:
n_out, n_in, _, _, _ = checkpoint[ln].size()
checkpoint[ln] = checkpoint[ln][:n_out // alpha * (alpha - 1), :n_in//beta,:,:,:]
model.load_state_dict(checkpoint,strict = False)
return model
def resnet101(alpha, beta):
"""Constructs a ResNet-101 model.
Args:
groups
"""
model = ResNet(Bottleneck, [3, 4, 23, 3], alpha,beta)
checkpoint = model_zoo.load_url(model_urls['resnet101'],model_dir='./pretrained/')
layer_name = list(checkpoint.keys())
for ln in layer_name:
if 'conv' in ln or 'downsample.0.weight' in ln:
checkpoint[ln] = checkpoint[ln].unsqueeze(2)
if 'conv2' in ln:
n_out, n_in, _, _, _ = checkpoint[ln].size()
checkpoint[ln] = checkpoint[ln][:n_out // alpha * (alpha - 1),:n_in//beta,:,:,:] #保留空间通道的预训练权重。
model.load_state_dict(checkpoint,strict = False)
return model
if __name__ == '__main__':
model = resnet50(8,2)
Input = torch.randn([1, 3, 8, 224, 224]) # N,C,T,W,H
out = model(Input)