Jie Xiao, Xueyang Fu, Man Zhou, Hongjian Liu, Zheng-Jun Zha
Paper: Link
- SIDD Google Drive
- SPA Google Drive
- GoPro Google Drive
python train/train_motiondeblur.py --arch ShuffleFormer_B --gpu '0,1,2,3' --batch_size 16 --train_ps 256 --train_dir GoPro/train --embed_dim 32 --warmup --nepoch 600 --save_dir save_dir --val_dir GoPro/val --train_workers 4 --weight_decay 0.0 --optimizer adam
# denoise
python test_sidd.py --arch ShuffleFormer_B --gpu '0' --val_dir data_path --pretrain_weights model_path --result_dir save_dir --repeat_num 16
# deblur&derain
python test.py --arch ShuffleFormer_B --gpu '0' --val_dir data_path --pretrain_weights model_path --result_dir save_dir --repeat_num 16
# evaluate: test/eval_spa.m eval_gopro.m(gopro&hide) eval_realblur.py(RealBlurR&RealBlurJ) evaluation code is borrowed from Restormer(https://github.com/swz30/Restormer)
We refer to Uformer and Restormer. Thanks for sharing.
Please contact us if there is any question([email protected]).