Xiaoccer / MobileFaceNet_Pytorch

MobileFaceNets: Efficient CNNs for Accurate Real-Time Face Verification on Mobile Devices
355 stars 88 forks source link

The dw Conv & the pw Conv? #7

Closed tt166111 closed 4 years ago

tt166111 commented 4 years ago

Hello~Thanks for your share. I notice that the pw conv doesn't have groups, but the dw conv has 'groups=inp * expansion' in your code, is it right ?

            #pw
            nn.Conv2d(inp, inp * expansion, 1, 1, 0, bias=False),
            #dw
            nn.Conv2d(inp * expansion, inp * expansion, 3, stride, 1, groups=inp * expansion, bias=False)