Multi-GPU Support

[sheshap]

Thank you so much for releasing the code.

I noticed the below error when I tried to training using 2 GPUs.

RuntimeError: arguments are located on different GPUs at /pytorch/aten/src/THC/generic/THCTensorMathBlas.cu:540

Please let us know if the code supports multi-GPU training?

[imankgoyal]

Hello,

Thanks foy your interest in our work. The multi-gpu training should work for PointNet++, PointNet, DGCNN. We have not tested it for RSCNN and SimpleView. Please share the exact command and config you were using.

Thanks,
Ankit

[sheshap]

Hi Ankit,

I am using below command.
python main.py --exp-config configs/dgcnn_simpleview_ce_run_1.yaml

The error at the last but 2 lines in file mv_utils.py

    points = torch.matmul(points, rot_mat)
    points = points - translation

If I comment these lines it works perfectly fine.

Thanks
Shivanand

[imankgoyal]

I checked the code. It looks like the current code does not support multi-gpu training for SimpleView. I edited my earlier comment. Please pull the commit (0612474) and it should work.

[sheshap]

Thank you. It worked.