shawnbit / unet-family Goto Github PK

I am trying to train the model using a custom dataset my dataset contains BSDS500 images. when I trying to train the model this error occurs:
`---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
in
24 original = original.to(device)
25 original = original.unsqueeze(0)
---> 26 out = model(original)

C:\Anaconda3\envs\torchgpu\lib\site-packages\torch\nn\modules\module.py in call(self, *input, **kwargs)
491 result = self._slow_forward(*input, **kwargs)
492 else:
--> 493 result = self.forward(*input, **kwargs)
494 for hook in self._forward_hooks.values():
495 hook_result = hook(self, input, result)

in forward(self, inputs)
51 up3 = self.up_concat3(up4,conv3) # 64128128
52 up2 = self.up_concat2(up3,conv2) # 32256256
---> 53 up1 = self.up_concat1(up2,conv1) # 16512512
54
55 final = self.final(up1)

C:\Anaconda3\envs\torchgpu\lib\site-packages\torch\nn\modules\module.py in call(self, *input, **kwargs)
491 result = self._slow_forward(*input, **kwargs)
492 else:
--> 493 result = self.forward(*input, **kwargs)
494 for hook in self._forward_hooks.values():
495 hook_result = hook(self, input, result)

in forward(self, high_feature, *low_feature)
53 outputs0 = self.up(high_feature)
54 for feature in low_feature:
---> 55 outputs0 = torch.cat([outputs0, feature], 1)
56 return self.conv(outputs0)

RuntimeError: invalid argument 0: Sizes of tensors must match except in dimension 1. Got 321 and 320 in dimension 2 at C:/w/1/s/tmp_conda_3.6_035809/conda/conda-bld/pytorch_1556683229598/work/aten/src\THC/generic/THCTensorMath.cu:71
`

Please help what is the problem and how can I solve it?
thanks a lot

如题

Can you please explain how to run this your code i.e. what is the input image to the model look like, how I can add the label for trained images, what is the procedure that needs to follow in training?

Hi,
I tried to train the model using a pixel wise criterion(L1 loss), my input image has 3 channels and since the final conv2d layer is using n_classes(by default is set to 2), for out_channels, the output has 2 channels and the criterion can not compare these images. Why should out_channels should be 2? Can it change to 3?
the code:
self.final = nn.Conv2d(filters[0], n_classes, 1)