train.md

Distributed / Multi GPU

CUDA_VISIBLE_DEVICES=0,1,2,3 \
torchrun --nproc_per_node=4 --master_port=4321 basicsr/train.py -opt options/train/GoPro/SegNAFNet.yml --launcher pytorch

If you try distributed, change batch_size_per_gpu and num_gpu in config yaml file. Batch_size = batch_size_per_gpu * num_gpu which we set 32.

Single GPU

python basicsr/train.py -opt options/train/GoPro/SegNAFNet.yml