Research map / Label-efficient and robust learning / Recognition and long-tail
Predicting Visual Overlap of Images Through Interpretable Non-Metric Box Embeddings
Anita Rau, Guillermo Garcia-Hernando, Danail Stoyanov, Gabriel J. Brostow, Daniyar Turmukhambetov
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use box embeddings with asymmetric overlap instead of symmetric vector embeddings | overlap prediction · Overlap prediction, Notre-Dame | L1 error: 0.244 -> 0.070 | Improved |
| use box embeddings instead of vectors | overlap prediction · Overlap prediction, Big Ben | accuracy (error < 0.1): 24.8% -> 82.6% | Improved |
| use box embeddings instead of vectors | overlap prediction · Overlap prediction, Venice | RMSE: 0.193 -> 0.112 | Improved |
| use box embeddings instead of vectors | overlap prediction · Overlap prediction, Florence | accuracy (error < 0.1): 76.5% -> 89.7% | Improved |
| use surface overlap instead of frustum overlap for vector embeddings (ablation) | retrieval for localization · 7Scenes localization, Stairs | median translation error (m): 1.82 -> 0.22 | Improved |
| use surface overlap instead of frustum overlap for vector embeddings (ablation) | retrieval for localization · 7Scenes localization, Chess | median rotation error (deg): 1.25 -> 1.19 | Improved |
| use box embeddings instead of symmetric vector embeddings | retrieval for localization · 7Scenes localization, Stairs | median translation error (m): 0.22 -> 0.50 | Hurt |
| use box embeddings instead of symmetric vector embeddings | retrieval for localization · 7Scenes localization, Chess | median translation error (m): 0.04 -> 0.05 | No clear effect |
| pre-scale images using predicted relative scale before matching | box embeddings · Notre-Dame and Big Ben pose estimation | rotation error | Improved |
Closest papers by meaning
- What VGGT Knows About Overlap: Probing Geometric Foundation Models for Co-Visibility (ECCV 2026 · similarity 0.87)
- Dense Correspondences across Scenes and Scales (TPAMI 2016 · similarity 0.87)
- Visiting the Invisible: Layer-by-Layer Completed Scene Decomposition (IJCV 2021 · similarity 0.87)
- Doppelgangers: Learning to Disambiguate Images of Similar Structures (ICCV 2023 · similarity 0.87)
- CrossOver: 3D Scene Cross-Modal Alignment (CVPR 2025 · similarity 0.86)
- Learning to Generate Multiple Objects from Dense and Occluded Layouts (NeurIPS 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
