pytorch f 函数

ShuffleNet是由2017年07月发布的轻量级网络，设计用于移动端设备，在MobileNet之后的网络架构。主要的创新点在于使用了分组卷积(group convolution)和通道打乱(channel shuffle)。

分组卷积和通道打乱

分组卷积

分组卷积最早由AlexNet中使用。由于当时的硬件资源有限，训练AlexNet时卷积操作不能全部放在同一个GPU处理，因此作者把特征图分给多个GPU分别进行处理，最后把特征图结果进行连接(concatenate)。

对于输入特征图$H\cdot W \cdot c_{1}$，将其按照channel-level分为$g$组，每组的通道数为$c_{1}/g$,因此卷积核的深度也要变为$c_{1}/g$，此时可认为有$g$个卷积核去对应处理一个特征图生成$g$个通道的特征图。若预期输出的特征图为通道数为$c_{2}$，则总卷积核数也是$c_{2}$。通道打乱

通道打乱并不是随机打乱，而是有规律地打乱。

(a)表示的是使用分组卷积处理未加打乱的特征图。每一个输出通道都对应组内的输入通道。(b)对组卷积GConv1处理后的特征图channel进行如上方式的打乱，此时大部分输出通道会对应不同组 的输入通道；(c )是与(b)的等价实现，将通道调整到合适的位置。

ShuffleNet单元

(a)是ResNet提出的bottleneck层，只是将3x3的卷积替换为depthwise(DW)卷积。(b)是1x1卷积变为pointwise分组卷积(GConv),并且加入了通道打乱。按照DW卷积的原论文，在3x3的DW后面不添加非线性因素。最后的点卷积用来匹配相加操作。( c)加入了stride=2去减少特征图尺寸，旁路加入平均池化层来维持相同的尺寸，最终在channel水平上连接特征图。

ShuffleNet网络结构

表格中的complexity表示FLOPs。stage2阶段的pointwise卷积不使用组卷机。对于一个bottleneck单元，对于输入维度为$c\times h\times w$的特征图，设bottleneck的channel数为$m$，输出channel数为$4m$。对于每一个stage的第一个bottleneck，对应图( c)，FLOPs的计算规则为：$h\times w\times c\times \frac{m}{g}+\frac{h}{2}\times \frac{w}{2}\times(9m+4m\times \frac{m}{g})\\$

对于其他bottleneck，输入和输出特征图的尺寸和channel数保持不变，所以为：$h\times w\times (cm/g+9m+cm/g)=hw(2cm/g+9m)$

'''ShuffleNet in PyTorch.

See the paper "ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices" for more details.
'''
import torch
import torch.nn as nn
import torch.nn.functional as F


class ShuffleBlock(nn.Module):
    def __init__(self, groups):
        super(ShuffleBlock, self).__init__()
        self.groups = groups

    def forward(self, x):
        '''Channel shuffle: [N,C,H,W] -> [N,g,C/g,H,W] -> [N,C/g,g,H,w] -> [N,C,H,W]'''
        N,C,H,W = x.size()
        g = self.groups
        # 维度变换之后必须要使用.contiguous()使得张量在内存连续之后才能调用view函数
        return x.view(N,g,int(C/g),H,W).permute(0,2,1,3,4).contiguous().view(N,C,H,W)


class Bottleneck(nn.Module):
    def __init__(self, in_planes, out_planes, stride, groups):
        super(Bottleneck, self).__init__()
        self.stride = stride

        # bottleneck层中间层的channel数变为输出channel数的1/4
        mid_planes = int(out_planes/4)


        g = 1 if in_planes==24 else groups
        # 作者提到不在stage2的第一个pointwise层使用组卷积,因为输入channel数量太少,只有24
        self.conv1 = nn.Conv2d(in_planes, mid_planes,
                               kernel_size=1, groups=g, bias=False)
        self.bn1 = nn.BatchNorm2d(mid_planes)
        self.shuffle1 = ShuffleBlock(groups=g)
        self.conv2 = nn.Conv2d(mid_planes, mid_planes,
                               kernel_size=3, stride=stride, padding=1,
                               groups=mid_planes, bias=False)
        self.bn2 = nn.BatchNorm2d(mid_planes)
        self.conv3 = nn.Conv2d(mid_planes, out_planes,
                               kernel_size=1, groups=groups, bias=False)
        self.bn3 = nn.BatchNorm2d(out_planes)

        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):
    def __init__(self, cfg):
        super(ShuffleNet, self).__init__()
        out_planes = cfg['out_planes']
        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_planes = 24
        self.layer1 = self._make_layer(out_planes[0], num_blocks[0], groups)
        self.layer2 = self._make_layer(out_planes[1], num_blocks[1], groups)
        self.layer3 = self._make_layer(out_planes[2], num_blocks[2], groups)
        self.linear = nn.Linear(out_planes[2], 10)

    def _make_layer(self, out_planes, num_blocks, groups):
        layers = []
        for i in range(num_blocks):
            if i == 0:
                layers.append(Bottleneck(self.in_planes,
                                         out_planes-self.in_planes,
                                         stride=2, groups=groups))
            else:
                layers.append(Bottleneck(self.in_planes,
                                         out_planes,
                                         stride=1, groups=groups))
            self.in_planes = out_planes
        return nn.Sequential(*layers)

    def forward(self, x):
        out = F.relu(self.bn1(self.conv1(x)))
        out = self.layer1(out)
        out = self.layer2(out)
        out = self.layer3(out)
        out = F.avg_pool2d(out, 4)
        out = out.view(out.size(0), -1)
        out = self.linear(out)
        return out


def ShuffleNetG2():
    cfg = {
        'out_planes': [200,400,800],
        'num_blocks': [4,8,4],
        'groups': 2
    }
    return ShuffleNet(cfg)

def ShuffleNetG3():
    cfg = {
        'out_planes': [240,480,960],
        'num_blocks': [4,8,4],
        'groups': 3
    }
    return ShuffleNet(cfg)


def test():
    net = ShuffleNetG2()
    x = torch.randn(1,3,32,32)
    y = net(x)
    print(y)

test()