README.md

Video Waterdrop Removal via Spatio-Temporal Fusion in Driving Scenes

Paper

This is the official PyTorch implementation. This works aims at removing various types of waterdrops for driving cars on rainy days. We also provide a large-scale synthetic dataset for the video waterdrop removal task.

Video Waterdrop Removal via Spatio-Temporal Fusion in Driving Scenes

Qiang Wen, Yue Wu, Qifeng Chen
The Hong Kong University of Science and Technology
IEEE International Conference on Robotics and Automation (ICRA), 2023

Requirements

• Pytorch 1.9
• OpenCV-Python

If conda has been installed, you can directly build the running environment via:

conda env create -f environment.yaml

An environment named "th" will be created.

Training

• Download the training dataset, and put it in dataset/;
• To train a model:

$ bash train.sh

You can also use the command tensorboard --logdir=runs to visually check the training results.

Testing

• Download the pretrained model, and put it in checkpoints_waterdrop/;
• Download the test dataset, and unzip it in dataset/;
• To test:

$ bash test.sh

You can choose the test on the synthetic dataset or real-world dataset by specifying --data_type

Citation

If you find this repository useful for your research, please cite the following work.

@inproceedings{wen2023video,
  title={Video Waterdrop Removal via Spatio-Temporal Fusion in Driving Scenes},
  author={Wen, Qiang and Wu, Yue and Chen, Qifeng},
  booktitle={International Conference on Robotics and Automation (ICRA)},
  year={2023},
  organization={IEEE}
}