replace argwhere with nonzero, thus could use lower version of torch

chaytonmin / Occupancy-MAE

Official implementation of our TIV'23 paper: Occupancy-MAE: Self-supervised Pre-training Large-scale LiDAR Point Clouds with Masked Occupancy Autoencoders

Apache License 2.0

251 stars 18 forks source link

replace argwhere with nonzero, thus could use lower version of torch #27

Closed SiyuanHuang95 closed 1 year ago

SiyuanHuang95 commented 1 year ago

Since most of the slurm training is with a100 or else, which is may not compatible with torch 1.13, so need to replace argwhere with nonzero.