Awesome

BrightFlow (WACV 2023)

This repository contains the official implementation of BrightFlow: Brightness-Change-Aware Unsupervised Learning of Optical Flow that has been published to the IEEE Winter Conference on Applications of Computer Vision (WACV) 2023.

Requirements

requirement.txt

Datasets

To train/evaluate BrightFlow or the baseline without BrightFlow, please download the required datasets:

• Sintel
• KITTI
• HD1K

Training

Baseline

sh script/train_baseline.sh

BrightFlow

sh script/train_brightflow.sh

Evaluation

The checkpoints of trained models are available here.

sh script/eval.sh

Acknowledgements

We thank authors of RAFT, GMA, SCV and SMURF for their great work and for sharing their code.

Citation

@inproceedings{marsal2023brightflow,
  title={BrightFlow: Brightness-Change-Aware Unsupervised Learning of Optical Flow},
  author={Marsal, Remi and Chabot, Florian and Loesch, Angelique and Sahbi, Hichem},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={2061--2070},
  year={2023}
}

License

This project is under the CeCILL license 2.1.