Enhanced Spatio-Temporal Interaction Learning for Video Deraining: A Faster and Better Framework

Kaihao Zhang, Dongxu Li, Wenhan Luo, Wenqi Ren, Wei Liu

Installation

To replicate the environment:

cd code
conda install --file requirements.txt

Training

Please first modify bash files accordingly with your data folder path.

cd code/run_scripts

(1) Train on NTU dataset: Put data under /$YOUR_ROOTPATH/derain/NTU-derain

cd code/run_scripts/
bash train_resnet18_5pic.sh

(2) Train on RainSys25 light dataset: Put data under /$YOUR_ROOTPATH/derain/RainSyn25

cd code/run_scripts/
bash train_resnet18_rainsys_light_5pic.sh

(3) Train on RainSys25 heavy dataset: Put data under /$YOUR_ROOTPATH/derain/RainSyn25

cd code/run_scripts/
bash train_resnet18_rainsys_heavy_5pic.sh

Testing with pre-trained weights

Please first modify bash files accordingly with your data folder path.

Download checkpoints and put in code/best_checkpoints (https://drive.google.com/drive/folders/19PSF-slyB_m_0dWWo-VRe_uvGOlDiCPf?usp=sharing)

(1) Test on NTU dataset:

cd code/run_scripts/
bash test_ntu_npic.sh

(2) Test on RainSys25 light dataset:

cd code/run_scripts/
bash test_light_npic.sh

(3) Test on RainSys25 heavy dataset:

cd code/run_scripts/
bash test_heavy_npic.sh

Citation

  @article{zhang2022enhanced,
    title={Enhanced Spatio-Temporal Interaction Learning for Video Deraining: A Faster and Better Framework},
    author={Zhang, Kaihao and Li, Dongxu and Luo, Wenhan and Ren, Wenqi and Liu, Wei},
    journal={IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)},
    year={2022}
  }