RuntimeError: CUDA out of memory.

[ArghyaChatterjee]

Hello, thanks for your work. I have seen your previous response regarding this question for training but I want this for testing purpose. Testing shouldn't take that much GPU memory. I have 3060 Ti Nvidia GPU with 6GB of Graphics memory and am testing on same klitti dataset with a batch size of 1.

arghya@arghya-Pulse-GL66-12UEK:~/PENet$ python main.py -b 1 -n pe --evaluate ~/PENet/model/pe.pth.tar --test --data-folder /home/arghya/kitti_depth/depth
Namespace(batch_size=1, convolutional_layer_encoding='xyz', cpu=False, criterion='l2', data_folder='/home/arghya/kitti_depth/depth', data_folder_rgb='data/dataset/kitti_raw', data_folder_save='data/dataset/kitti_depth/submit_test/', dilation_rate=2, epochs=100, evaluate='/home/arghya/PENet/model/pe.pth.tar', freeze_backbone=False, input='rgbd', jitter=0.1, lr=0.001, network_model='pe', not_random_crop=False, print_freq=10, random_crop_height=320, random_crop_width=1216, rank_metric='rmse', result='../results', resume='', start_epoch=0, start_epoch_bias=0, test=True, use_d=True, use_g=True, use_rgb=True, val='select', val_h=352, val_w=1216, weight_decay=1e-06, workers=4)
=> using 'cuda' for computation.
=> loading checkpoint '/home/arghya/PENet/model/pe.pth.tar' ... Completed.
=> creating model and optimizer ... => checkpoint state loaded.
=> creating source code backup ...
=> finished creating source code backup.
=> logger created.
Traceback (most recent call last):
  File "main.py", line 474, in <module>
    main()
  File "main.py", line 386, in main
    iterate("test_completion", args, test_loader, model, None, logger, 0)
  File "main.py", line 216, in iterate
    pred = model(batch_data)
  File "/home/arghya/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/arghya/PENet/model.py", line 503, in forward
    depth5 = self.CSPN5_s2(guide5_s2, depth5, coarse_depth)
  File "/home/arghya/.local/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1102, in _call_impl
    return forward_call(*input, **kwargs)
  File "/home/arghya/PENet/basic.py", line 267, in forward
    output = torch.einsum('ijk,ijk->ik', (input_im2col, kernel))
  File "/home/arghya/.local/lib/python3.8/site-packages/torch/functional.py", line 325, in einsum
    return einsum(equation, *_operands)
  File "/home/arghya/.local/lib/python3.8/site-packages/torch/functional.py", line 327, in einsum
    return _VF.einsum(equation, operands)  # type: ignore[attr-defined]
RuntimeError: CUDA out of memory. Tried to allocate 42.00 MiB (GPU 0; 5.81 GiB total capacity; 3.75 GiB already allocated; 4.38 MiB free; 3.81 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF