360° Optical Flow using Tangent Images

Video | Paper

This repository contains the source code for BMVC 2021 paper "360° Optical Flow using Tangent Images".

In this project root folder run pip install -r requirements.txt to install the dependency.

• Run python main.py to estimate the 360° optical flow with demo data.

• Run test_replica360.py to estimate the 360° optical flow on the whole Replica 360° dataset and quantitative evaluate the result.

• Replica 360°

Please download the Replica 360° optical flow data from link.

The mask file xxxx_mask_pano.png indicate the unavailable pixels. The value 0 pixel's RGB, depth and optical flow are unavailable, the value 1 pixels are available.

• OmniPhotos

Please download the 360° OmniPhotos dataset from OmniPhotos webpage.

If you think our code useful, please consider citing our paper:

@InProceedings{yuan2021panoof,
  title={360° Optical Flow using Tangent Images},
  author={Yuan, Mingze and Christian, Richardt},
  booktitle = {Proceedings of the 32nd British Machine Vision Conference (BMVC)},
  year={2021}
}

We thank the reviewers for their valuable feedback that has helped us improve our paper.

This work was supported by an EPSRC-UKRI Innovation Fellowship (EP/S001050/1) and RCUK grant CAMERA (EP/M023281/1, EP/T022523/1).

This repository is released under the MIT license.