Booooooooooo
commented
3 years ago I have found that in
https://github.com/jacquelinelala/GFN/blob/3b80f530d9a04964fb80300b08267ae2d4c78753/train_GFN_4x.py#L102
epoch_loss += msewill accumulate gradients of all iterations and finally cause CUDA out of memory. Useepoch_loss += mse.item()instead will solve this problem.
I am facing issue RuntimeError: CUDA error: out of memory. I had follow the instructions.
Traceback (most recent call last): File "test_GFN_4x.py", line 130, in
model_test(model)
File "test_GFN_4x.py", line 100, in model_test
test(testloader, model, criterion, SR_dir)
File "test_GFN_4x.py", line 77, in test
[lr_deblur, sr] = model(LR_Blur, gated_Tensor, test_Tensor)
File "D:\InstalledApps\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 477, in call
result = self.forward(*input, kwargs)
File "D:\Work\VInventTechWork\Biop.ai\Deblurry\GFN\networks\GFN_4x.py", line 216, in forward
recon_out = self.reconstructMoudle(fusion_feature)
File "D:\InstalledApps\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 477, in call
result = self.forward(*input, *kwargs)
File "D:\InstalledApps\Anaconda3\lib\site-packages\torch\nn\modules\container.py", line 91, in forward
input = module(input)
File "D:\InstalledApps\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 477, in call
result = self.forward(input, kwargs)
File "D:\Work\VInventTechWork\Biop.ai\Deblurry\GFN\networks\GFN_4x.py", line 186, in forward
pixelshuffle1 = self.relu1(self.pixelShuffle1(con1))
File "D:\InstalledApps\Anaconda3\lib\site-packages\torch\nn\modules\module.py", line 477, in call
result = self.forward(*input, **kwargs)
File "D:\InstalledApps\Anaconda3\lib\site-packages\torch\nn\modules\pixelshuffle.py", line 40, in forward
return F.pixel_shuffle(input, self.upscale_factor)
File "D:\InstalledApps\Anaconda3\lib\site-packages\torch\nn\functional.py", line 1844, in pixel_shuffle
shuffle_out = input_view.permute(0, 1, 4, 2, 5, 3).contiguous()
RuntimeError: CUDA error: out of memory