Fast-SCNN-pytorch
Fast-SCNN-pytorch copied to clipboard
Tramac
assert target.dim() == 3 error
I got a error in loss.py line 54. I print my target size is (2,768,768,3), but the code shows it should be dim 3. my predict size is (2,3,768,768). I use my own dataset, picture annotation like cityscapes. then I get this error. I don't know what happen.
Hi @midtom How did you resolve this issue? I am facing it at the moment when I use my own model for training. Thanks in advance.
@dronefreak ,@midtom did you resolve this issue? I have the same problem,can you give me some advice? thank you!
@dronefreak ,@midtom did you resolve this issue? I have the same problem,can you give me some advice? thank you!
Please confirm whether the label is single channel.
This is essentially a dimension mismatch issue, looks like your target size is in channels_last format whereas your prediction is in channels_first. Reshaping either of the two tensors will solve your issue. Also, please make sure that your both target and predictions are single channel tensors.
@Tramac,@dronefreak,i meet anthor problem:
when i reshape both my target size and image size are channels_first,an error was reported below:
Traceback (most recent call last): File "train.py", line 208, in <module> trainer.train() File "train.py", line 130, in train for i, (images, targets) in enumerate(self.train_loader):
File "/home/liyanzhou/anaconda3/envs/py36torch10_fastscnn/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 560, in __next__ batch = self.collate_fn([self.dataset[i] for i in indices])
File "/home/liyanzhou/anaconda3/envs/py36torch10_fastscnn/lib/python3.6/site-packages/torch/utils/data/dataloader.py", line 560, in <listcomp> batch = self.collate_fn([self.dataset[i] for i in indices])
File "/home/liyanzhou/pycharm_project/demo/Fast-SCNN-pytorch-master/data_loader/segbase.py", line 50, in __getitem__ img = self.transform(img)
File "/home/liyanzhou/anaconda3/envs/py36torch10_fastscnn/lib/python3.6/site-packages/torchvision/transforms/transforms.py", line 61, in __call__ img = t(img)
File "/home/liyanzhou/anaconda3/envs/py36torch10_fastscnn/lib/python3.6/site-packages/torchvision/transforms/transforms.py", line 164, in __call__ return F.normalize(tensor, self.mean, self.std, self.inplace)
File "/home/liyanzhou/anaconda3/envs/py36torch10_fastscnn/lib/python3.6/site-packages/torchvision/transforms/functional.py", line 208, in normalize tensor.sub_(mean[:, None, None]).div_(std[:, None, None])
RuntimeError: The size of tensor a (768) must match the size of tensor b (3) at non-singleton dimension 0
the image size and target size are transform transpose img shape: (1, 768, 768) transform transpose mask shape: (1, 768, 768)
how to resolve it?i need some advice,thank you!
Oh you would need to expand the dimensions using torch.unsqueeze(), this will add a 4th dim to your tensor (batch size), which is expected by your pytorch model. Input tensors to the model should have 4 dimensions, [batch_size, channels, height, width].
Wonderful!Maybe I read the images was wrong! I should read them with RGB,but I read them with single channel.Thank you!
Now the network can run,but the loss is high loss =0.1887. The learning rate is 0.001.
By the way,the shape of outputs and target are torch.Size([4, 2, 768, 768]) torch.Size([768, 768]),is this normal?
@Tramac @dronefreak
Okay so that means you have a batch size of 4 and perhaps you are doing a binary (2-class) classification? Target size seems a bit off, as there should be the num_channels and batch_dim as well. Ideally your label should have the same size as that of your output, otherwise you would run into potential dimension mismatch issues, while calculating your losses.