Skip to content
/ SPDNet Public
forked from Joyies/SPDNet

Structure-Preserving Deraining with Residue Channel Prior Guidance (ICCV2021)

Notifications You must be signed in to change notification settings

MIVRC/SPDNet

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

39 Commits
 
 
 
 

Repository files navigation

SPDNet

Structure-Preserving Deraining with Residue Channel Prior Guidance (ICCV2021) arXiv GitHub Stars

Requirements

  • Linux Platform
  • NVIDIA GPU + CUDA CuDNN
  • PyTorch == 0.4.1
  • torchvision0.2.0
  • pytorch_wavelets
  • Python3.6.0
  • imageio2.5.0
  • numpy1.14.0
  • opencv-python
  • scikit-image0.13.0
  • tqdm4.32.2
  • scipy1.2.1
  • matplotlib3.1.1
  • ipython7.6.1
  • h5py2.10.0

Training

  1. Modify data path in src/data/rainheavy.py and src/data/rainheavytest.py
    datapath/data/***.png
    datapath/label/***.png
  2. Begining training:
$ cd ./src/
$ python main.py --save spdnet --model spdnet --scale 2 --epochs 300 --batch_size 16 --patch_size 128 --data_train RainHeavy --n_threads 0 --data_test RainHeavyTest --data_range 1-1800/1-200 --loss 1*MSE  --save_results --lr 5e-4 --n_feats 32 --n_resblocks 3

Test

The pre-trained model can be available at google drive: https://drive.google.com/drive/folders/1ylON5AkJVayoypOXDaUEkYd76LtMF-lB?usp=sharing.

$ cd ./src/
$ python main.py --data_test RainHeavyTest  --ext img --scale 2  --data_range 1-1800/1-200 --pre_train ../experiment/spdnet/model/model_best.pt --model spdnet --test_only --save_results --save SPDNet_test

All PSNR and SSIM results are computed by using this Matlab code, based on Y channel of YCbCr space.

Datasets

Rain200H: 1800 training pairs and 200 testing pairs
Rain200L: 1800 training pairs and 200 testing pairs
Rain800: 700 training pairs and 100 testing pairs
Rain1200: 12000 traing paris and 1200 testing pairs
SPA-Data: 638492 training pairs and 1000 testing pairs

Acknowledgement

Code borrows from RCDNet. Thanks for sharing !

About

Structure-Preserving Deraining with Residue Channel Prior Guidance (ICCV2021)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%