torch-mesh-isect icon indicating copy to clipboard operation
torch-mesh-isect copied to clipboard

Published 20 hours ago • verified publisher icon vchoutas

CUDA 11 support

Open tengyu-liu opened this issue 3 years ago • 1 comments

Hi,

I'm encountering some strange errors when using mesh_intersection on 3090 GPU, CUDA 11.1, python 3.8.8, pytorch 1.8.1.

I encountered a CUDA out of memory error when running examples/batch_smpl_untangle.py using suggested parameters. If I add a line val += np.random.normal(0, 0.1, val.shape) between examples/batch_smpl_untangle.py:129 and 130, coll_idx can sometimes have a negative length in loss.py:224.

It looks like that the problem is with CUDA11. Is it possible for you to support CUDA 11?

Thank you very much!

tengyu-liu avatar Jun 18 '21 04:06 tengyu-liu

I have tried on CUDA 11.3 + python3.7 + pytorch 1.11~ it works well

wangsen1312 avatar May 02 '22 00:05 wangsen1312