Tie Liu, Mai Xu, Zulin Wang
[Paper Link] (ICME'2019 Oral)
Highlights
- We construct a large-scale dataset of 316 synthetic rain videos.
- We propose a novel deep learning architecture with two-stream ConvLSTM based subnets for the joint tasks of rain streak detection and removal on videos..
A large-scale video database for rain removal (LasVR). The videos of our database contain a wide range of content categories, e.g., animal, nature landscapes, human scenes, action sports, man-made object and so forth. Additionally, the rendered streaks vary from light drizzling to heavy rain storm and vertical rain to slash line. The training, validation and testing sets consist of 261, 27 and 28 rain videos.
Download LasVR-Database:
[Dropbox] [BaiduYun] (password:bwg0)
Main code is in main.py. We cropped training videos into numerous 64x64x9 cubes, and the code for feeding training data should be done by users.
Environment
- Python
- TensorFlow
- TFLearn
If you find this work helpful, please cite:
@inproceedings{liu2019removing,
title={Removing rain in videos: a large-scale database and a two-stream ConvLSTM approach},
author={Liu, Tie and Xu, Mai and Wang, Zulin},
booktitle={2019 IEEE International Conference on Multimedia and Expo (ICME)},
pages={664--669},
year={2019},
organization={IEEE}
}
[email protected], [email protected] (Tie Liu)
You can use, redistribute, and adapt the material for non-commercial purposes, as long as you give appropriate credit by citing our paper and indicating any changes that you've made.