search

Harry-Zhi / semantic_nerf

The implementation of "In-Place Scene Labelling and Understanding with Implicit Scene Representation" [ICCV 2021].
Other
418 stars 54 forks source link

Can it work with multi-GPUs? #29

Closed XinyueZ closed 1 year ago

XinyueZ commented 1 year ago

in config just see "gpu". can we use more than one to speed up training?

Harry-Zhi commented 1 year ago

Hi @XinyueZ ,

Yes, defintely it could work with more GPUS to support more efficient training as well as dense samples and rays. In my current implementation I only use a single GPU but its extention to multi-GPUs could be directly applied.

XinyueZ commented 1 year ago

@Harry-Zhi Thank you. Can you give a little detail on how to work with multi-GPUs within this repo?

Harry-Zhi commented 1 year ago

@Harry-Zhi Thank you. Can you give a little detail on how to work with multi-GPUs within this repo?

Hi. I think shifting to multi-GPUs and DDP could be straightforward if you follow the instructions here https://pytorch.org/tutorials/beginner/ddp_series_multigpu.html