Trinity-Net

Trinity-Net: Gradient-Guided Swin Transformer-based Remote Sensing Image Dehazing and Beyond.

This is the code of the implementation of the Trinity-Net.

Download code from Baidu Cloud: https://pan.baidu.com/s/1uz4fInYLAzbqhN2nqgQznA?pwd=1004 key: 1004 or Google Drive: https://drive.google.com/file/d/1CMjoDwt8xMjZ5wpjvNnWoY5Cf9XZmQH5/view?usp=sharing

Requirement

ubuntu, torch==1.8.1+cu111, torchvision==0.9.1+cu111, tensorboardX==2.5.1.

Training

1. Put the training data to corresponding folders (hazy image to ./data/train_data/input, GT to ./data/train_data/target)
2. Hyperparameter (./Enh_opt)
3. Python Enh_train.py

Testing

1. Put the testing data to corresponding folders (hazy image to ./data/test_data/input, GT to ./data/test_data/target, GT for full-reference evaluation, such as PSNR and SSIM)
2. Python Enh_eval.py
3. Find the result in corresponding folder (./checkpoints/XX/test_results)

Remote Sensing Image Dehazing Dataset (RSID)

Download RSID from Baidu Cloud: https://pan.baidu.com/s/1zzk1KiKJHnZPHg4BV5U7dA?pwd=1004 key: 1004 or Google Drive: https://drive.google.com/file/d/1FC7oSkGTthjHl2sKN-yGrKhssgV0QB4F/view?usp=sharing

Natural Image Dehazing Dataset (NID)

Download NID from Baidu Cloud: https://pan.baidu.com/s/1bvXiWE3kVH_xhISL_SJ6xA?pwd=1004 key: 1004 or Google Drive: https://drive.google.com/file/d/1vyGsFDaV9uVMO4Qeg1dRYitbIDYSC_eX/view?usp=sharing

Contact Us

If you have any questions, please contact us (chikaichen@mail.nwpu.edu.cn).