Input channel mismatch about ffc

[huameinan219]

I meet the same problem if ration_gin is not zero!

I meet the same problem if ration_gin is not zero!
我也是遇到的这个问题，然后我参考了文件ffc_resnet.py中实现BisicBlock类的做法，我使用三层ffc来代替原来的卷积层。
self.ffclayers = nn.Sequential(FFC(channels, channels, 3, 0, 0.75, stride=1, padding=1, dilation=1, groups=1, bias=False, enable_lfu=False), FFC(channels, channels, 3, 0.75, 0.75, stride=1, padding=1, dilation=1, groups=1, bias=False, enable_lfu=False), FFC(channels, self.out_channel, 3, 0.75, 0, stride=1, padding=1, dilation=1, groups=1, bias=False, enable_lfu=False), ConcatTupleLayer(),)
这里的ConcatTupleLayer()参考的 https://github.com/saic-mdal/lama

class ConcatTupleLayer(nn.Module): def forward(self, x): assert isinstance(x, tuple) x_l, x_g = x assert torch.is_tensor(x_l) or torch.is_tensor(x_g) if not torch.is_tensor(x_g): return x_l return torch.cat(x, dim=1)
class ConcatTupleLayer(nn.Module): def forward(self, x): assert isinstance(x, tuple) x_l, x_g = x assert torch.is_tensor(x_l) or torch.is_tensor(x_g) if not torch.is_tensor(x_g): return x_l return torch.cat(x, dim=1)

我在一个人脸修复任务中的开头的卷积层应用了快速傅里叶卷积，效果好像还彳亍。

[ZeliangM]

I meet the same problem if ration_gin is not zero!

[ZeliangM]

I meet the same problem if ration_gin is not zero!

I meet the same problem if ration_gin is not zero! 我也是遇到的这个问题，然后我参考了文件ffc_resnet.py中实现BisicBlock类的做法，我使用三层ffc来代替原来的卷积层。 self.ffclayers = nn.Sequential(FFC(channels, channels, 3, 0, 0.75, stride=1, padding=1, dilation=1, groups=1, bias=False, enable_lfu=False), FFC(channels, channels, 3, 0.75, 0.75, stride=1, padding=1, dilation=1, groups=1, bias=False, enable_lfu=False), FFC(channels, self.out_channel, 3, 0.75, 0, stride=1, padding=1, dilation=1, groups=1, bias=False, enable_lfu=False), ConcatTupleLayer(),) 这里的ConcatTupleLayer()参考的 https://github.com/saic-mdal/lama

class ConcatTupleLayer(nn.Module): def forward(self, x): assert isinstance(x, tuple) x_l, x_g = x assert torch.is_tensor(x_l) or torch.is_tensor(x_g) if not torch.is_tensor(x_g): return x_l return torch.cat(x, dim=1) class ConcatTupleLayer(nn.Module): def forward(self, x): assert isinstance(x, tuple) x_l, x_g = x assert torch.is_tensor(x_l) or torch.is_tensor(x_g) if not torch.is_tensor(x_g): return x_l return torch.cat(x, dim=1)

我在一个人脸修复任务中的开头的卷积层应用了快速傅里叶卷积，效果好像还彳亍。

感谢您耐心回复，对我解决bug很有帮助。希望未来能有机会和您进一步交流。