search

lizhaoliu-Lec / CPCM

This is the official repo for Contextual Point Cloud Modeling for Weakly-supervised Point Cloud Semantic Segmentation (ICCV 23).
MIT License
34 stars 3 forks source link

question about multi-GPUs training #5

Closed mooncakehub closed 12 months ago

mooncakehub commented 1 year ago

thank you for your excellent work. because my sigle gpu's memory isn't enough, I want to know if your code can be modified use multi-gpus to train.

lizhaoliu-Lec commented 1 year ago

The code supports multi GPU training, controlled by the flag CUDA_VISIBLE_DEVICES.

For now, the default settings for S3DIS and ScanNet V2 are 2 point-clouds per-GPU.

To reduce the GPU memory consumption, just set DATA.batch_size 1. I hope the batch size of 1 can be supported by your GPU.

mooncakehub commented 1 year ago

thank you for your reply. I have several 2080ti gpus ,but each of them has 12G memory.So I want to use two gpus to train S3DIS. Because DATA.batch_size 1 may not perform well as your paper said. I haven't tried many gpus parallels, so I don't know if it's feasible