search

sanghyun-son / EDSR-PyTorch

PyTorch version of the paper 'Enhanced Deep Residual Networks for Single Image Super-Resolution' (CVPRW 2017)
MIT License
2.41k stars 666 forks source link

Training EDSR is VERY SLOW!! #318

Open Suanmd opened 2 years ago

Suanmd commented 2 years ago

I tried to run CUDA_VISIBLE_DEVICES=0 python main.py --model EDSR --scale 2 --save edsr_x2 --n_resblocks 32 --n_feats 256 --res_scale 0.1 --reset --ext sep_reset on a single Tesla V100, but the training is very slow. This script takes several hours to run 10 epochs and consumes 27251M of memory. Is this normal? Thanks!

我尝试在单张Tesla V100上运行 CUDA_VISIBLE_DEVICES=0 python main.py --model EDSR --scale 2 --save edsr_x2 --n_resblocks 32 --n_feats 256 --res_scale 0.1 --reset --ext sep_reset 但是训练速度非常缓慢。该命令需要几个小时才能跑10个epoch，并且内存占用是27251M。 这正常吗？谢谢！

msx-123 commented 2 years ago

I also have the same problem, may I ask who can reply to me after solving it? Thank you

ulfailliyina commented 2 years ago

hello,do you know? how to fix this https://github.com/sanghyun-son/EDSR-PyTorch/issues/319#issue-996722398 thankyou

HolmesShuan commented 2 years ago

@Suanmd @msx-123 FYI.

CUDA_VISIBLE_DEVICES=0 python main.py --model EDSR --scale 2 --save edsr_x2 --n_resblocks 32 --n_feats 256 --res_scale 0.1 --reset --ext sep_reset --patch_size 96
# Multi-GPUs 
# CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python main.py --model EDSR --scale 2 --save edsr_x2 --n_resblocks 32 --n_feats 256 --res_scale 0.1 --reset --ext sep_reset --patch_size 96 --n_GPUs 8 --chop