The for loop of optimizer.step() and loss.backward(retain_graph=True) #1
Description
Traceback (most recent call last):
File "F:\Aseq_DIP-main\denoising_main.py", line 167, in
loss.backward(retain_graph=True)
File "C:\Users\admin\anaconda3\envs\Hyperspectral\Lib\site-packages\torch\tensor.py", line 221, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph)
File "C:\Users\admin\anaconda3\envs\Hyperspectral\Lib\site-packages\torch\autograd_init_.py", line 132, in backward
allow_unreachable=True) # allow_unreachable flag
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [3, 64, 1, 1]] is at version 2; expected version 1 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).
This is the problem when i run the author's denoising code.
Is there some problem with the order of optimizer.step() and loss.backward(retain_graph=True), when I try to remove the for loop in the optimizer and switch the order of the two, the program works fine, is this in line with the original paper in Neural IPS?