深度可分离卷积(Depthwise Separable Convolution)和分组卷积(Group Convolution)的理解，相互关系及PyTorch实现_separable convolution pytorch实现

1. 分组卷积(Group Convolution)

分组卷积最早出现在AlexNet中，如下图所示。在CNN发展初期，GPU资源不足以满足训练任务的要求，因此，Hinton采用了多GPU训练的策略，每个GPU完成一部分卷积，最后把多个GPU的卷积结果进行融合。

这里提出一个小小的问题给大家思考：如上图所示，input Features 是12,将其分为3个组，每组4个Features map,那么output Feature maps 的数量可以是任意的吗，可以是1吗？

2. 深度可分离卷积(Depthwise Separable Convolution)

3. PyTorch实现

Pytorch是2017年推出的深度学习框架，不同于Tensorflow基于静态图的模型搭建方式，PyTorch是完全动态的框架，推出以来很快成为AI研究人员的热门选择并受到推崇。（介绍到此结束）

在PyTorch中，实现二维卷积是通过nn.Conv2d实现的，这个函数是非常强大的，其功能不仅仅是实现常规卷积，通过合理的参数选择就可以实现分组卷积、空洞卷积。API的官方介绍如下：

CLASS torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True)
 
 - stride: controls the stride for the cross-correlation, a single number or a tuple.
 - padding: controls the amount of implicit zero-paddings on both sides for padding number of points for each dimension.
 - dilation: controls the spacing between the kernel points; also known as the à trous algorithm. It is harder to describe, but this link has a nice visualization of what dilation does.
 - groups: controls the connections between inputs and outputs. in_channels and out_channels must both be divisible by groups. For example,
		At groups=1, all inputs are convolved to all outputs.
		At groups=2, the operation becomes equivalent to having two conv layers side by side, each seeing half the input channels, and producing half the output channels, and both subsequently concatenated.
		At groups= in_channels, each input channel is convolved with its own set of filters.

3.1 分组卷积

分组卷积只需要对nn.Conv2d中的groups参数进行设置即可，表示需要分的组数，groups的默认值为1，即进行常规卷积。以下是实现分组卷积的代码：

class CSDN_Tem(nn.Module):
    def __init__(self, in_ch, out_ch, groups):
        super(CSDN_Tem, self).__init__()
        self.conv = nn.Conv2d(
            in_channels=in_ch,
            out_channels=out_ch,
            kernel_size=3,
            stride=1,
            padding=1,
            groups=groups
        )
 
    def forward(self, input):
        out = self.conv(input)
        return out

通过以下代码对该模型进行测试，设定输入特征图通道数为16，输出特征图通道数为64，分组数目为4：

conv = CSDN_Tem(16, 64, 4)
print(summary(conv, (16, 128, 128), batch_size=1))

控制台输出为：

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1          [1, 64, 128, 128]           2,368
================================================================
Total params: 2,368
Trainable params: 2,368
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 1.00
Forward/backward pass size (MB): 8.00
Params size (MB): 0.01
Estimated Total Size (MB): 9.01
----------------------------------------------------------------

这一分组卷积过程所需参数为2368个，其中包含了偏置(Bias).

3.2 深度可分离卷积

深度可分离卷积的PyTorch代码如下：

class CSDN_Tem(nn.Module):
    def __init__(self, in_ch, out_ch):
        super(CSDN_Tem, self).__init__()
        self.depth_conv = nn.Conv2d(
            in_channels=in_ch,
            out_channels=in_ch,
            kernel_size=3,
            stride=1,
            padding=1,
            groups=in_ch
        )
        self.point_conv = nn.Conv2d(
            in_channels=in_ch,
            out_channels=out_ch,
            kernel_size=1,
            stride=1,
            padding=0,
            groups=1
        )
 
    def forward(self, input):
        out = self.depth_conv(input)
        out = self.point_conv(out)
        return out

采用和分组卷积相同的输入和输出通道数，测试代码如下：

conv = depth_conv(16,64)
print(summary(conv,(16,128,128),batch_size=1))

控制台输出结果为：

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1          [1, 16, 128, 128]             160
            Conv2d-2          [1, 64, 128, 128]           1,088
================================================================
Total params: 1,248
Trainable params: 1,248
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 1.00
Forward/backward pass size (MB): 10.00
Params size (MB): 0.00
Estimated Total Size (MB): 11.00

深度可分离卷积实现相同的操作仅需1248个参数。

如有疑问，欢迎留言！

参考

1. https://www.cnblogs.com/shine-lee/p/10243114.html
2. https://blog.csdn.net/luoluonuoyasuolong/article/details/81750190
3. https://blog.csdn.net/luoluonuoyasuolong/article/details/81750190
4. https://pytorch.org/docs/stable/nn.html?highlight=conv2#torch.nn.Conv2d