CUDA out of memory. about robust-lane-detection HOT 4 CLOSED

The releaased code is running on Pytorch0.4.0. If adapted to Pytorch1.1.0 or above, optimizer.step()' should be before lr_scheduler.step()'.
For the problem of out of memory, you may have to set a smaller batch size.

from robust-lane-detection.

Thank you for your answer.
I have set the batch size to 3, but it still didn't work.
It really confused me that a 16GB GPU was out of memory.
As I know, Pytorch won't take multi-GPU itself.
Can you please tell me how did you do that?
Thank you so much!

from robust-lane-detection.

I found I forgot to resize the image causing to the memory problem.
However, I face another problem.

Traceback (most recent call last):
File "c:/Users/10806337/Desktop/PortableGit/Projects/Robust-Lane-Detection/LaneDetectionCode/train.py", line 122, in
train(args, epoch, model, train_loader, device, optimizer, criterion)
File "c:/Users/10806337/Desktop/PortableGit/Projects/Robust-Lane-Detection/LaneDetectionCode/train.py", line 21, in train
loss.backward()
File "C:\Users\10806337.conda\envs\RobustLaneDetection\lib\site-packages\torch\tensor.py", line 118, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph)
File "C:\Users\10806337.conda\envs\RobustLaneDetection\lib\site-packages\torch\autograd_init_.py", line 93, in backward
allow_unreachable=True) # allow_unreachable flag
RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn

Do you have any idea of it?

from robust-lane-detection.

I have solved my problem.
I accidentally add torch.no_grad(): before the training loop.
But I found one bug in the code that is in train():
loss = criterion(output, target)
should be
loss = criterion(output[0], target)

from robust-lane-detection.