File size: 10,400 Bytes
bdbd148
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
'''
ShuffleNetV2 in PyTorch.

ShuffleNetV2是ShuffleNet的改进版本,通过实验总结出了四个高效网络设计的实用准则:
1. 输入输出通道数相等时计算量最小
2. 过度使用组卷积会增加MAC(内存访问代价)
3. 网络碎片化会降低并行度
4. Element-wise操作不可忽视

主要改进:
1. 通道分离(Channel Split)替代组卷积
2. 重新设计了基本单元,使输入输出通道数相等
3. 每个阶段使用不同的通道数配置
4. 简化了下采样模块的设计

Reference:
[1] Ningning Ma, Xiangyu Zhang, Hai-Tao Zheng, Jian Sun
    ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design. ECCV 2018.
'''
import torch
import torch.nn as nn
import torch.nn.functional as F


class ShuffleBlock(nn.Module):
    """通道重排模块
    
    通过重新排列通道的顺序来实现不同特征的信息交流。
    
    Args:
        groups (int): 分组数量,默认为2
    """
    def __init__(self, groups=2):
        super(ShuffleBlock, self).__init__()
        self.groups = groups

    def forward(self, x):
        """通道重排的前向传播
        
        步骤:
        1. [N,C,H,W] -> [N,g,C/g,H,W]  # 重塑为g组
        2. [N,g,C/g,H,W] -> [N,C/g,g,H,W]  # 转置g维度
        3. [N,C/g,g,H,W] -> [N,C,H,W]  # 重塑回原始形状
        
        Args:
            x: 输入张量,[N,C,H,W]
            
        Returns:
            out: 通道重排后的张量,[N,C,H,W]
        """
        N, C, H, W = x.size()
        g = self.groups
        return x.view(N, g, C//g, H, W).permute(0, 2, 1, 3, 4).reshape(N, C, H, W)


class SplitBlock(nn.Module):
    """通道分离模块
    
    将输入特征图按比例分成两部分。
    
    Args:
        ratio (float): 分离比例,默认为0.5
    """
    def __init__(self, ratio):
        super(SplitBlock, self).__init__()
        self.ratio = ratio

    def forward(self, x):
        """通道分离的前向传播
        
        Args:
            x: 输入张量,[N,C,H,W]
            
        Returns:
            tuple: 分离后的两个张量,[N,C1,H,W]和[N,C2,H,W]
        """
        c = int(x.size(1) * self.ratio)
        return x[:, :c, :, :], x[:, c:, :, :]


class BasicBlock(nn.Module):
    """ShuffleNetV2的基本模块
    
    结构:
    x -------|-----------------|
      |      |                |
      |      1x1 Conv         |
      |      3x3 DWConv       |
      |      1x1 Conv         |
      |                       |
      |------------------Concat
                          |
                      Channel Shuffle
                          
    Args:
        in_channels (int): 输入通道数
        split_ratio (float): 通道分离比例,默认为0.5
    """
    def __init__(self, in_channels, split_ratio=0.5):
        super(BasicBlock, self).__init__()
        self.split = SplitBlock(split_ratio)
        in_channels = int(in_channels * split_ratio)
        
        # 主分支
        self.conv1 = nn.Conv2d(in_channels, in_channels,
                              kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm2d(in_channels)
        
        self.conv2 = nn.Conv2d(in_channels, in_channels,
                              kernel_size=3, stride=1, padding=1, 
                              groups=in_channels, bias=False)
        self.bn2 = nn.BatchNorm2d(in_channels)
        
        self.conv3 = nn.Conv2d(in_channels, in_channels,
                              kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm2d(in_channels)
        
        self.shuffle = ShuffleBlock()

    def forward(self, x):
        # 通道分离
        x1, x2 = self.split(x)
        
        # 主分支
        out = F.relu(self.bn1(self.conv1(x2)))
        out = self.bn2(self.conv2(out))
        out = F.relu(self.bn3(self.conv3(out)))
        
        # 拼接并重排
        out = torch.cat([x1, out], 1)
        out = self.shuffle(out)
        return out


class DownBlock(nn.Module):
    """下采样模块
    
    结构:
           3x3 DWConv(s=2)     1x1 Conv
    x -----> 1x1 Conv          3x3 DWConv(s=2)
                               1x1 Conv
                                  |
                              Concat
                                  |
                          Channel Shuffle
                          
    Args:
        in_channels (int): 输入通道数
        out_channels (int): 输出通道数
    """
    def __init__(self, in_channels, out_channels):
        super(DownBlock, self).__init__()
        mid_channels = out_channels // 2
        
        # 左分支
        self.branch1 = nn.Sequential(
            # 3x3深度可分离卷积,步长为2
            nn.Conv2d(in_channels, in_channels,
                     kernel_size=3, stride=2, padding=1, 
                     groups=in_channels, bias=False),
            nn.BatchNorm2d(in_channels),
            # 1x1卷积
            nn.Conv2d(in_channels, mid_channels,
                     kernel_size=1, bias=False),
            nn.BatchNorm2d(mid_channels)
        )
        
        # 右分支
        self.branch2 = nn.Sequential(
            # 1x1卷积
            nn.Conv2d(in_channels, mid_channels,
                     kernel_size=1, bias=False),
            nn.BatchNorm2d(mid_channels),
            # 3x3深度可分离卷积,步长为2
            nn.Conv2d(mid_channels, mid_channels,
                     kernel_size=3, stride=2, padding=1,
                     groups=mid_channels, bias=False),
            nn.BatchNorm2d(mid_channels),
            # 1x1卷积
            nn.Conv2d(mid_channels, mid_channels,
                     kernel_size=1, bias=False),
            nn.BatchNorm2d(mid_channels)
        )
        
        self.shuffle = ShuffleBlock()

    def forward(self, x):
        # 左分支
        out1 = self.branch1(x)
        
        # 右分支
        out2 = self.branch2(x)
        
        # 拼接并重排
        out = torch.cat([out1, out2], 1)
        out = self.shuffle(out)
        return out


class ShuffleNetV2(nn.Module):
    """ShuffleNetV2模型
    
    网络结构:
    1. 一个卷积层进行特征提取
    2. 三个阶段,每个阶段包含多个基本块和一个下采样块
    3. 最后一个卷积层
    4. 平均池化和全连接层进行分类
    
    Args:
        net_size (float): 网络大小系数,可选0.5/1.0/1.5/2.0
    """
    def __init__(self, net_size):
        super(ShuffleNetV2, self).__init__()
        out_channels = configs[net_size]['out_channels']
        num_blocks = configs[net_size]['num_blocks']

        # 第一层卷积
        self.conv1 = nn.Conv2d(3, 24, kernel_size=3,
                              stride=1, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(24)
        self.in_channels = 24
        
        # 三个阶段
        self.layer1 = self._make_layer(out_channels[0], num_blocks[0])
        self.layer2 = self._make_layer(out_channels[1], num_blocks[1])
        self.layer3 = self._make_layer(out_channels[2], num_blocks[2])
        
        # 最后的1x1卷积
        self.conv2 = nn.Conv2d(out_channels[2], out_channels[3],
                              kernel_size=1, stride=1, padding=0, bias=False)
        self.bn2 = nn.BatchNorm2d(out_channels[3])
        
        # 分类层
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.classifier = nn.Linear(out_channels[3], 10)
        
        # 初始化权重
        self._initialize_weights()

    def _make_layer(self, out_channels, num_blocks):
        """构建一个阶段
        
        Args:
            out_channels (int): 输出通道数
            num_blocks (int): 基本块的数量
            
        Returns:
            nn.Sequential: 一个阶段的层序列
        """
        layers = [DownBlock(self.in_channels, out_channels)]
        for i in range(num_blocks):
            layers.append(BasicBlock(out_channels))
            self.in_channels = out_channels
        return nn.Sequential(*layers)

    def forward(self, x):
        """前向传播
        
        Args:
            x: 输入张量,[N,3,32,32]
            
        Returns:
            out: 输出张量,[N,num_classes]
        """
        # 特征提取
        out = F.relu(self.bn1(self.conv1(x)))
        
        # 三个阶段
        out = self.layer1(out)
        out = self.layer2(out)
        out = self.layer3(out)
        
        # 最后的特征提取
        out = F.relu(self.bn2(self.conv2(out)))
        
        # 分类
        out = self.avg_pool(out)
        out = out.view(out.size(0), -1)
        out = self.classifier(out)
        return out
    
    def _initialize_weights(self):
        """初始化模型权重
        
        采用kaiming初始化方法:
        - 卷积层权重采用kaiming_normal_初始化
        - BN层参数采用常数初始化
        - 线性层采用正态分布初始化
        """
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
                if m.bias is not None:
                    nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.Linear):
                nn.init.normal_(m.weight, 0, 0.01)
                nn.init.constant_(m.bias, 0)


# 不同大小的网络配置
configs = {
    0.5: {
        'out_channels': (48, 96, 192, 1024),
        'num_blocks': (3, 7, 3)
    },
    1.0: {
        'out_channels': (116, 232, 464, 1024),
        'num_blocks': (3, 7, 3)
    },
    1.5: {
        'out_channels': (176, 352, 704, 1024),
        'num_blocks': (3, 7, 3)
    },
    2.0: {
        'out_channels': (224, 488, 976, 2048),
        'num_blocks': (3, 7, 3)
    }
}


def test():
    """测试函数"""
    # 创建模型
    net = ShuffleNetV2(net_size=0.5)
    print('Model Structure:')
    print(net)
    
    # 测试前向传播
    x = torch.randn(1,3,32,32)
    y = net(x)
    print('\nInput Shape:', x.shape)
    print('Output Shape:', y.shape)
    
    # 打印模型信息
    from torchinfo import summary
    device = 'cuda' if torch.cuda.is_available() else 'cpu'
    net = net.to(device)
    summary(net, (1,3,32,32))


if __name__ == '__main__':
    test()