在做depthwise convolution的时候,遇到了groups参数,初步理解:
out_channels中,groups个共用一组weight。
example:
self.dw = nn.Conv2d_bn_ReLU(cin_channels, cin_channels, kernel_size=3, stride=stride, groups=c*in_channels)
为depthwise卷积。
08-06
08-06
08-06