|
''' |
|
ShuffleNet in PyTorch. |
|
|
|
ShuffleNet是一个专门为移动设备设计的高效卷积神经网络。其主要创新点在于使用了两个新操作: |
|
1. 逐点组卷积(pointwise group convolution) |
|
2. 通道重排(channel shuffle) |
|
这两个操作大大降低了计算复杂度,同时保持了良好的准确率。 |
|
|
|
主要特点: |
|
1. 使用组卷积减少参数量和计算量 |
|
2. 使用通道重排操作使不同组之间的信息可以流通 |
|
3. 使用深度可分离卷积进一步降低计算复杂度 |
|
4. 设计了多个计算复杂度版本以适应不同的设备 |
|
|
|
Reference: |
|
[1] Xiangyu Zhang, Xinyu Zhou, Mengxiao Lin, Jian Sun |
|
ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices. CVPR 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组之间的信息交流。 |
|
|
|
Args: |
|
groups (int): 分组数量 |
|
""" |
|
def __init__(self, groups): |
|
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 Bottleneck(nn.Module): |
|
"""ShuffleNet的基本模块 |
|
|
|
结构: |
|
x -> 1x1 GConv -> BN -> Shuffle -> 3x3 DWConv -> BN -> 1x1 GConv -> BN -> (+) -> ReLU |
|
|---------------------| |
|
|
|
Args: |
|
in_channels (int): 输入通道数 |
|
out_channels (int): 输出通道数 |
|
stride (int): 步长,用于下采样 |
|
groups (int): 组卷积的分组数 |
|
""" |
|
def __init__(self, in_channels, out_channels, stride, groups): |
|
super(Bottleneck, self).__init__() |
|
self.stride = stride |
|
|
|
|
|
mid_channels = out_channels // 4 |
|
g = 1 if in_channels == 24 else groups |
|
|
|
|
|
self.conv1 = nn.Conv2d(in_channels, mid_channels, |
|
kernel_size=1, groups=g, bias=False) |
|
self.bn1 = nn.BatchNorm2d(mid_channels) |
|
self.shuffle1 = ShuffleBlock(groups=g) |
|
|
|
|
|
self.conv2 = nn.Conv2d(mid_channels, mid_channels, |
|
kernel_size=3, stride=stride, padding=1, |
|
groups=mid_channels, bias=False) |
|
self.bn2 = nn.BatchNorm2d(mid_channels) |
|
|
|
|
|
self.conv3 = nn.Conv2d(mid_channels, out_channels, |
|
kernel_size=1, groups=groups, bias=False) |
|
self.bn3 = nn.BatchNorm2d(out_channels) |
|
|
|
|
|
self.shortcut = nn.Sequential() |
|
if stride == 2: |
|
self.shortcut = nn.Sequential( |
|
nn.AvgPool2d(3, stride=2, padding=1) |
|
) |
|
|
|
def forward(self, x): |
|
|
|
out = F.relu(self.bn1(self.conv1(x))) |
|
out = self.shuffle1(out) |
|
out = F.relu(self.bn2(self.conv2(out))) |
|
out = self.bn3(self.conv3(out)) |
|
|
|
|
|
res = self.shortcut(x) |
|
|
|
|
|
out = F.relu(torch.cat([out, res], 1)) if self.stride == 2 else F.relu(out + res) |
|
return out |
|
|
|
|
|
class ShuffleNet(nn.Module): |
|
"""ShuffleNet模型 |
|
|
|
网络结构: |
|
1. 一个卷积层进行特征提取 |
|
2. 三个阶段,每个阶段包含多个带重排的残差块 |
|
3. 平均池化和全连接层进行分类 |
|
|
|
Args: |
|
cfg (dict): 配置字典,包含: |
|
- out_channels (list): 每个阶段的输出通道数 |
|
- num_blocks (list): 每个阶段的块数 |
|
- groups (int): 组卷积的分组数 |
|
""" |
|
def __init__(self, cfg): |
|
super(ShuffleNet, self).__init__() |
|
out_channels = cfg['out_channels'] |
|
num_blocks = cfg['num_blocks'] |
|
groups = cfg['groups'] |
|
|
|
|
|
self.conv1 = nn.Conv2d(3, 24, kernel_size=1, bias=False) |
|
self.bn1 = nn.BatchNorm2d(24) |
|
self.in_channels = 24 |
|
|
|
|
|
self.layer1 = self._make_layer(out_channels[0], num_blocks[0], groups) |
|
self.layer2 = self._make_layer(out_channels[1], num_blocks[1], groups) |
|
self.layer3 = self._make_layer(out_channels[2], num_blocks[2], groups) |
|
|
|
|
|
self.avg_pool = nn.AdaptiveAvgPool2d(1) |
|
self.classifier = nn.Linear(out_channels[2], 10) |
|
|
|
|
|
self._initialize_weights() |
|
|
|
def _make_layer(self, out_channels, num_blocks, groups): |
|
"""构建ShuffleNet的一个阶段 |
|
|
|
Args: |
|
out_channels (int): 输出通道数 |
|
num_blocks (int): 块的数量 |
|
groups (int): 分组数 |
|
|
|
Returns: |
|
nn.Sequential: 一个阶段的层序列 |
|
""" |
|
layers = [] |
|
for i in range(num_blocks): |
|
stride = 2 if i == 0 else 1 |
|
cat_channels = self.in_channels if i == 0 else 0 |
|
layers.append( |
|
Bottleneck( |
|
self.in_channels, |
|
out_channels - cat_channels, |
|
stride=stride, |
|
groups=groups |
|
) |
|
) |
|
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] |
|
""" |
|
|
|
out = F.relu(self.bn1(self.conv1(x))) |
|
|
|
|
|
out = self.layer1(out) |
|
out = self.layer2(out) |
|
out = self.layer3(out) |
|
|
|
|
|
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) |
|
|
|
|
|
def ShuffleNetG2(): |
|
"""返回groups=2的ShuffleNet模型""" |
|
cfg = { |
|
'out_channels': [200,400,800], |
|
'num_blocks': [4,8,4], |
|
'groups': 2 |
|
} |
|
return ShuffleNet(cfg) |
|
|
|
|
|
def ShuffleNetG3(): |
|
"""返回groups=3的ShuffleNet模型""" |
|
cfg = { |
|
'out_channels': [240,480,960], |
|
'num_blocks': [4,8,4], |
|
'groups': 3 |
|
} |
|
return ShuffleNet(cfg) |
|
|
|
|
|
def test(): |
|
"""测试函数""" |
|
|
|
net = ShuffleNetG2() |
|
print('Model Structure:') |
|
print(net) |
|
|
|
|
|
x = torch.randn(1,3,32,32) |
|
y = net(x) |
|
print('\nInput Shape:', x.shape) |
|
print('Output Shape:', y.shape) |
|
|
|
|
|
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() |