Sonata: Self-Supervised Learning of Reliable Point Representations
Xiaoyang Wu, Daniel DeTone, Duncan Frost, Tianwei Shen, Chris Xie, Nan Yang, Jakob Engel, Richard Newcombe, Hengshuang Zhao, Julian Straub · CVPR 2025 · 2025
- Status
- To read
- Topics
- backbones
- Feeds
- First paper · height sampling
What they did
Identifies a "geometric shortcut" that makes self-supervised point representations collapse to low-level spatial features, and avoids it by obscuring spatial information and relying more on input features, giving representations that work under linear probing.
My note
No note yet.
Code
- Repository
- facebookresearch/sonata
- Framework
- PyTorch
- Licence
- Apache-2.0
- My status
- Not tried
BibTeX
@inproceedings{wu2025-sonata,
title = {{Sonata: Self-Supervised Learning of Reliable Point Representations}},
author = {Xiaoyang Wu and Daniel DeTone and Duncan Frost and Tianwei Shen and Chris Xie and Nan Yang and Jakob Engel and Richard Newcombe and Hengshuang Zhao and Julian Straub},
booktitle = {CVPR 2025},
year = {2025},
doi = {10.1109/CVPR52734.2025.02067},
eprint = {2503.16429},
archivePrefix = {arXiv}
}