This library provides tools to use the ensemble self-training method to train pose detectors in a self-supervised manner.
If you find this paper helpful, please cite:
@article{ShiTalak23rss-ensemble,
title = {A Correct-and-Certify Approach to Self-Supervise Object Pose Estimators via Ensemble Self-Training},
author = {J. Shi and R. Talak and D. Maggio and L. Carlone},
journal = {Robotics: Science and Systems (RSS)},
pdf={https://arxiv.org/pdf/2302.06019.pdf},
year = {2023}
}
More documentation and trained models will be available soon.
This repo relies heavily on CosyPose. Huge thanks to the original authors for releasing their code open-source.