Cross-View Completion Models are Zero-shot Correspondence Estimators
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2024
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866916519865221120 |
|---|---|
| author | An, Honggyu Kim, Jinhyeon Park, Seonghoon Jung, Jaewoo Han, Jisang Hong, Sunghwan Kim, Seungryong |
| author_facet | An, Honggyu Kim, Jinhyeon Park, Seonghoon Jung, Jaewoo Han, Jisang Hong, Sunghwan Kim, Seungryong |
| contents | In this work, we explore new perspectives on cross-view completion learning by drawing an analogy to self-supervised correspondence learning. Through our analysis, we demonstrate that the cross-attention map within cross-view completion models captures correspondence more effectively than other correlations derived from encoder or decoder features. We verify the effectiveness of the cross-attention map by evaluating on both zero-shot matching and learning-based geometric matching and multi-frame depth estimation. Project page is available at https://cvlab-kaist.github.io/ZeroCo/. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_09072 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Cross-View Completion Models are Zero-shot Correspondence Estimators An, Honggyu Kim, Jinhyeon Park, Seonghoon Jung, Jaewoo Han, Jisang Hong, Sunghwan Kim, Seungryong Computer Vision and Pattern Recognition In this work, we explore new perspectives on cross-view completion learning by drawing an analogy to self-supervised correspondence learning. Through our analysis, we demonstrate that the cross-attention map within cross-view completion models captures correspondence more effectively than other correlations derived from encoder or decoder features. We verify the effectiveness of the cross-attention map by evaluating on both zero-shot matching and learning-based geometric matching and multi-frame depth estimation. Project page is available at https://cvlab-kaist.github.io/ZeroCo/. |
| title | Cross-View Completion Models are Zero-shot Correspondence Estimators |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2412.09072 |