使用resnet50作为backbone时候提示形状不匹配

ZhengPeng7 / BiRefNet

[CAAI AIR'24] Bilateral Reference for High-Resolution Dichotomous Image Segmentation

https://www.birefnet.top

MIT License

1.09k stars 84 forks source link

使用resnet50作为backbone时候提示形状不匹配 #19

Open wangwuwu opened 5 months ago

wangwuwu commented 5 months ago

x4 = self.squeeze_module(x4)： RuntimeError: Given groups=1, weight of size [64, 3712, 3, 3], expected input[4, 1856, 128, 128] to have 3712 channels, but got 1856 channels instea

ZhengPeng7 commented 5 months ago

Hi, sorry for that. There are indeed still some bugs in using ConvNets, I'll try to spare time to fix them. You can first try the vision transformer backbones.