windflow_light

Perform optical flow inference on geostationary satellite images from a pretrained RAFT model.

Install

conda env create -f environment_cpu.yml

Usage

See python predict.py for a basic example.

Citation

Vandal, T., Duffy, K., McCarty, W., Sewnath, A., & Nemani, R. (2022). Dense feature tracking of atmospheric winds with deep optical flow, Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining.

Acknowledgements

External packages and flownet code was used from: https://github.com/celynw/flownet2-pytorch/
Funded by NASA ROSES Earth Science Research from Geostationary Satellite Program (2020-2023)