Skip to content

Latest commit

 

History

History
30 lines (24 loc) · 1.48 KB

README.md

File metadata and controls

30 lines (24 loc) · 1.48 KB

HomoFormer: Homogenized Transformer for Image Shadow Removal (CVPR 2024)

Jie Xiao, Xueyang Fu, Yurui Zhu, Dong Li, Jie Huang, Kai Zhu, Zheng-Jun Zha

Method Overview

Pretrained Model

Pre-trained weights can be found at this page.

train

# SRD(HomoFormer) with two 1080Ti GPUs
python train.py --save_dir your_save_dir --train_dir SRD_Train_DIR --val_dir SRD_Test --optimizer adam --env _SRD --arch HomoFormer --batch_size 4 --train_ps 384 

test

# SRD(HomoFormer)
python test_M.py --input_dir datapath --weights pretrained_weights --arch HomoFormer --cal_metrics  --result_dir savedir --save_images --gpus 0 --repeat 1 --base_repeat 1
SRD(HomoFormer+)
python test_M.py --input_dir datapath --weights pretrained_weights --arch HomoFormer --cal_metrics  --result_dir savedir --save_images --gpus 0 --repeat 16 --base_repeat 4
ISTDP(HomoFormer)
python test_M.py --input_dir datapath --weights pretrained_weights --arch HomoFormer --cal_metrics  --result_dir savedir --save_images --gpus 0 --repeat 1 --base_repeat 1 --plus
ISTDP(HomoFormer+)
python test_M.py --input_dir datapath --weights pretrained_weights --arch HomoFormer --cal_metrics  --result_dir savedir --save_images --gpus 0 --repeat 16 --base_repeat 4 --plus

Contact

Please contact us if there is any question(ustchbxj@mail.ustc.edu.cn).