Neural Outlier Rejection for Self-Supervised Keypoint Learning
Code coming soon.
Overview
- IO-Net: A novel proxy task for the self-supervision of keypoint description.
- KeyPointNet: An improved keypoint-network architecture that is especially amenable to robust keypoint detection and description.
Over-fitting Examples
- Toy example:
- TRI example:
Qualatitive Results
- Illumination Cases:
- Perspective Cases:
- Rotation Cases:
Citation
Please use the following citation when referencing our work:
@inproceedings{
tang2020neural,
title={Neural Outlier Rejection for Self-Supervised Keypoint Learning},
author={Jiexiong Tang and Hanme Kim and Vitor Guizilini and Sudeep Pillai and Rares Ambrus},
booktitle={International Conference on Learning Representations},
year={2020}
}