This is the code for Multi-view object pose estimation from correspondence distributions and epipolar geometry ICRA 2023.
The tless-model used for the experiments is provided under releases. In contrast to original surfemb, this model is trained on amodal mask crops, used in EpiSurfEmb.
For more information, see the surfemb paper.
Citing EpiSurfEmb:
@inproceedings{haugaard2023multi,
title={Multi-view object pose estimation from correspondence distributions and epipolar geometry},
author={Haugaard, Rasmus Laurvig and Iversen, Thorbjorn Mosekjaer},
booktitle={2023 IEEE International Conference on Robotics and Automation (ICRA)},
pages={1786--1792},
year={2023},
organization={IEEE}
}Citing SurfEmb:
@inproceedings{haugaard2022surfemb,
title={Surfemb: Dense and continuous correspondence distributions for object pose estimation with learnt surface embeddings},
author={Haugaard, Rasmus Laurvig and Buch, Anders Glent},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={6749--6758},
year={2022}
}