RuntimeError: CUDA error: out of memory for "DRN-D-105" while testing about drn

[taesungp]

It's because the new pytorch deprecated volatile, which was used to disable gradient recording. The new recommended way is using torch.no_grad().

In the last line of segment.py, wrap main() with with torch.no_grad():

if __name__ == "__main__":
    with torch.no_grad():
        main()

[jwzhi]

same issue

[lyxbyr]

this is bug???

[lyxbyr]

same issue,could you solve it?

[raven38]

Because the crop_size argument is disabled while testing. The argument is enabled while only training.
Please refer

drn/segment.py

Lines 632 to 640 in d75db2e

	dataset = SegListMS(data_dir, phase, transforms.Compose([
	transforms.ToTensor(),
	normalize,
	]), scales, list_dir=args.list_dir)
	else:
	dataset = SegList(data_dir, phase, transforms.Compose([
	transforms.ToTensor(),
	normalize,
	]), list_dir=args.list_dir, out_name=True)

and

drn/segment.py

Lines 360 to 383 in d75db2e

	t = []
	if args.random_rotate > 0:
	t.append(transforms.RandomRotate(args.random_rotate))
	if args.random_scale > 0:
	t.append(transforms.RandomScale(args.random_scale))
	t.extend([transforms.RandomCrop(crop_size),
	transforms.RandomHorizontalFlip(),
	transforms.ToTensor(),
	normalize])
	train_loader = torch.utils.data.DataLoader(
	SegList(data_dir, 'train', transforms.Compose(t),
	list_dir=args.list_dir),
	batch_size=batch_size, shuffle=True, num_workers=num_workers,
	pin_memory=True, drop_last=True
	)
	val_loader = torch.utils.data.DataLoader(
	SegList(data_dir, 'val', transforms.Compose([
	transforms.RandomCrop(crop_size),
	transforms.ToTensor(),
	normalize,
	]), list_dir=args.list_dir),
	batch_size=batch_size, shuffle=False, num_workers=num_workers,
	pin_memory=True, drop_last=True
	)