Escaping Plato's Cave: Towards the Alignment of 3D and Text Latent Spaces
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915323106557952 |
|---|---|
| author | Hadgi, Souhail Moschella, Luca Santilli, Andrea Gomez, Diego Huang, Qixing Rodolà, Emanuele Melzi, Simone Ovsjanikov, Maks |
| author_facet | Hadgi, Souhail Moschella, Luca Santilli, Andrea Gomez, Diego Huang, Qixing Rodolà, Emanuele Melzi, Simone Ovsjanikov, Maks |
| contents | Recent works have shown that, when trained at scale, uni-modal 2D vision and text encoders converge to learned features that share remarkable structural properties, despite arising from different representations. However, the role of 3D encoders with respect to other modalities remains unexplored. Furthermore, existing 3D foundation models that leverage large datasets are typically trained with explicit alignment objectives with respect to frozen encoders from other representations. In this work, we investigate the possibility of a posteriori alignment of representations obtained from uni-modal 3D encoders compared to text-based feature spaces. We show that naive post-training feature alignment of uni-modal text and 3D encoders results in limited performance. We then focus on extracting subspaces of the corresponding feature spaces and discover that by projecting learned representations onto well-chosen lower-dimensional subspaces the quality of alignment becomes significantly higher, leading to improved accuracy on matching and retrieval tasks. Our analysis further sheds light on the nature of these shared subspaces, which roughly separate between semantic and geometric data representations. Overall, ours is the first work that helps to establish a baseline for post-training alignment of 3D uni-modal and text feature spaces, and helps to highlight both the shared and unique properties of 3D data compared to other representations. Our code and weights are available at https://github.com/Souhail-01/3d-text-alignment |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2503_05283 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Escaping Plato's Cave: Towards the Alignment of 3D and Text Latent Spaces Hadgi, Souhail Moschella, Luca Santilli, Andrea Gomez, Diego Huang, Qixing Rodolà, Emanuele Melzi, Simone Ovsjanikov, Maks Computer Vision and Pattern Recognition Recent works have shown that, when trained at scale, uni-modal 2D vision and text encoders converge to learned features that share remarkable structural properties, despite arising from different representations. However, the role of 3D encoders with respect to other modalities remains unexplored. Furthermore, existing 3D foundation models that leverage large datasets are typically trained with explicit alignment objectives with respect to frozen encoders from other representations. In this work, we investigate the possibility of a posteriori alignment of representations obtained from uni-modal 3D encoders compared to text-based feature spaces. We show that naive post-training feature alignment of uni-modal text and 3D encoders results in limited performance. We then focus on extracting subspaces of the corresponding feature spaces and discover that by projecting learned representations onto well-chosen lower-dimensional subspaces the quality of alignment becomes significantly higher, leading to improved accuracy on matching and retrieval tasks. Our analysis further sheds light on the nature of these shared subspaces, which roughly separate between semantic and geometric data representations. Overall, ours is the first work that helps to establish a baseline for post-training alignment of 3D uni-modal and text feature spaces, and helps to highlight both the shared and unique properties of 3D data compared to other representations. Our code and weights are available at https://github.com/Souhail-01/3d-text-alignment |
| title | Escaping Plato's Cave: Towards the Alignment of 3D and Text Latent Spaces |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2503.05283 |