Mmdetection: how to assign specific gpus?

[chenxu93]

I have 4 gpus eg,0,1,2,3 but I want to use 2,3 to train my models. What should I do to train my model successfully?

but the way,I use CUDA_VISIBLE_DIVICES=2,3 xxxxx --gpus 2, and CUDA_VISIBLE_DIVICES=0,1,2,3 xxxx --gpus 2 both are not useless.

[ZhihuaGao]

It works for me in dist_train.sh...

#!/usr/bin/env bash

PYTHON=${PYTHON:-"python"}

CUDA_VISIBLE_DEVICES=4,5,6,7 $PYTHON -m torch.distributed.launch --nproc_per_node=$2 $(dirname "$0")/train.py $1 --launcher pytorch ${@:3}

[chenxu93]

thank you very much, do you put these 3 commands in a shell scrtip?

[ZhihuaGao]

Yes

[chenxu93]

ok, I will try, thank you !!!

[ZhihuaGao]

No thank you!

[chenxu93]

HI, @AresGao in your commands I do not know how to assign the config file, in the README FILE I find author say distribute train use the command like:./tools/dist_train.sh [optional arguments]. The command we can assign the config file. Can you make an explanation

[chenxu93]

Oh, I reference https://github.com/facebookresearch/maskrcnn-benchmark, and use following command to train successfully.

export NGPUS=2
CUDA_VISIBLE_DEVICES=2,3 python -m torch.distributed.launch --nproc_per_node=$NGPUS tools/train.py configs/faster_rcnn_r101_fpn_1x.py --gpus 2