ip-derain / seidnet

SEIDNet

we propose SEIDNet equipped with the generative Status Estimation and Information Decoupling for rain removal.

Requirements

• python 3.7.11
• pytorch 1.8.0
• CUDA Version 11.4
• opencv-python 4.4.0.44
• scikit-image 0.17.2

Datasets

Single image rain removal

• Rain100L-old_version https://github.com/nnUyi/DerainZoo/blob/master/DerainDatasets.md
• Rain100H-old_version https://github.com/nnUyi/DerainZoo/blob/master/DerainDatasets.md
• Rain1400 https://xueyangfu.github.io/projects/cvpr2017.html
• SPA https://stevewongv.github.io/derain-project.html
• Rain13k https://github.com/swz30/MPRNet

Single image snow removal

• Snow100K https://goo.gl/BrRc3U

Single image haze removal

• ITS&OTS https://sites.google.com/site/boyilics/website-builder/reside

Single image shadow removal

• ISTD Dataset https://drive.google.com/file/d/1I0qw-65KBA6np8vIZzO6oeiOvcDBttAY/view

Pretrained models

Coming soon

Train

sh train_SEIDNet.sh

Test

sh test_SEIDNet.sh

Results

Single image rain removal

We compare SEIDNet with state-of-the-art methods on the test sets of Rain100H, Rain100L, Rain1400, SPA and Rain13K. The performances are reported in terms of PSNR and SSIM.

We compare SEIDNet with state-of-the-art methods on the test sets of Rain13K. The performances are reported in terms of PSNR and SSIM.

Single image snow removal

We compare SEIDNet with state-of-the-art methods on the test sets of Snow100K. The performances are reported in terms of PSNR and SSIM.

Single image haze removal

We compare SEIDNet with other methods on ITS&OTS. The results are listed in terms of PSNR and SSIM.

Single image shadow removal

We compare SEIDNet with other methods on the test set of ISTD. The per- formances are reported in term of RMSE.

Bibtex