Source model from https://github.com/leonmakise/NBNet_Pytorch
Paper https://arxiv.org/abs/2012.15028
This repo. supports training on multiple GPUs.
Train
CUDA_VISIBLE_DEVICES=0,1 python train.py --noise_dir ../image/noise/ --gt_dir ../image/gt/ --image_size 512 --batch_size 16 --save_every 100 --loss_every 10 -nw 4 -c -ckpt nbnet --restart```
If no --restart
, the train process would be resumed.
python test_custom_mat.py -n /mnt/vinai/SIDD/ValidationNoisyBlocksSrgb.mat -g /mnt/vinai/SIDD/ValidationGtBlocksSrgb.mat -c -ckpt mir_kpn -m KPN
--save_img
to save image result