Self-Guided Dual Attention Network for UDC Image Restoration

Requirements

• Python 3.5
• PyTorch 1.1.0
• CUDA 10.2
• Scipy 1.1.0
• PIL 5.2.0
• Opencv 3.4.2

Image Patch Pairs Generation

• Change "gt_folder", "input_folder", "output_gt_folder", "output_input_folder" folder path in img_patch_gen.py
• Change "patch_size" and "stride" in img_patch_gen.py
• Run img_patch_gen.py

Train

• Change the parameters
• Run pyhton train.py

Test

Track 2

• python test.py --valid_file (test mat file) --saving_img_root (output png file root) --saving_mat_dir (result mat zip saving root) --load_name (model name)

Pretrain model

• https://drive.google.com/drive/folders/1EAAycawumKEVaZVI2JoRX32tC1Er13Ar?usp=sharing
• https://portland-my.sharepoint.com/:f:/g/personal/kinwailau6-c_ad_cityu_edu_hk/EqXe_Rk9aFNKvwrBOOmToWkBv-JjCpbvWaDPZyiJ5TC24w?e=8FZcTl