Multiview Equivariance Improves 3D Correspondence Understanding with Minimal Feature Finetuning

Published in ICLR, 2025

Yang You, Yixin Li, Congyue Deng, Yue Wang, Leonidas Guibas

arXiv PDF Code Project page BibTeX

Abstract

This work evaluates and improves the 3D awareness of Vision Transformer (ViT)-based models, showing that enhancing 3D equivariance in their semantic embeddings leads to better performance in tasks like pose estimation and tracking. The authors propose a simple finetuning strategy based on 3D correspondences, demonstrating substantial improvements with minimal finetuning on a single object.

BibTeX

@inproceedings{you2025multiview,
  title     = {Multiview Equivariance Improves 3D Correspondence Understanding with Minimal Feature Finetuning},
  author    = {Yang You and Yixin Li and Congyue Deng and Yue Wang and Leonidas Guibas},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year      = {2025},
  eprint    = {2411.19458},
  archivePrefix = {arXiv}
}

Recommended citation: You, Y., Li, Y., Deng, C., Wang, Y., & Guibas, L. (2025). Multiview Equivariance Improves 3D Correspondence Understanding with Minimal Feature Finetuning. In International Conference on Learning Representations (ICLR).