When converting the weight file to an onnx file, there are two identity nodes participating in the BN layer. Have you encountered this problem

XuJiacong / PIDNet

This is the official repository for our recent work: PIDNet

MIT License

596 stars 109 forks source link

When converting the weight file to an onnx file, there are two identity nodes participating in the BN layer. Have you encountered this problem #98

Open Daniu-jiao opened 4 months ago

Daniu-jiao commented 4 months ago

This mistake in class PAPPM last shortcut moduel.

out = self.compression(torch.cat(x_list, 1)) + self.shortcut(x)

self.shortcut = nn.Sequential( BatchNorm(inplanes, momentum=bn_mom), nn.ReLU(inplace=True), nn.Conv2d(inplanes, outplanes, kernel_size=1, bias=False), )