分组卷积(Group Convolution)与深度可分离卷积(Depthwise Separable Convolution)
常规卷积输入input[1,3,64,64],kernel[3,4,3,3],stride=1,padding=0,输出output[1,4,62,62]importtorchx=torch.rand(1,3,64,64)m=torch.nn.Conv2d(3,4,kernel_size=3,stride=1)y=m(x)分组卷积如上图,左边是常规卷积,假设input_features[1,12,