Research map / 3D vision / Depth and stereo
A Light Touch Approach to Teaching Transformers Multi-View Geometry
Yash Bhalgat, João F. Henriques, Andrew Zisserman
This paper tried a common change: Add positional encoding
Use explicit Epipolar Positional Encoding instead of implicit Epipolar Loss · CO3D-Retrieve full images · R@1: 90.57 -> 88.53 Hurt
What 646 accepted papers report for this change
- 525 improved
- 58 no clear effect
- 7 mixed
- 56 hurt
- Learning-Augmented Scalable Linear Assignment Problem Optimization via Neural Dual Warm-Starts ICML 2026 · add positional encodings to features (d=13 -> d=21) · Block-structured, N=4096 · speedupImproved
- Mesh Based Simulations with Spatial and Temporal awareness ICML 2026 · use learned positional embedding, learnable relative bias or weighted adjacency instead of RoPE · Aneurysm · RMSEHurt
- Sparse ActionGen: Accelerating Diffusion Policy with Real-time Pruning ICML 2026 · add sinusoidal positional encoding to the pruner · Robomimic Tool Hang · success rate (%): 28 -> 54Improved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
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 |
|---|---|---|---|
| Add Epipolar Loss to RRT + R50 | RRT + R50 · CO3D-Retrieve full images | R@1: 89.20 -> 90.57 | Improved |
| Add Epipolar Loss to RRT + R50 | RRT + R50 · CO3D-Retrieve masked backgrounds | R@1: 83.28 -> 85.07 | Improved |
| Add Epipolar Loss to RRT + R50 | RRT + R50 · Stanford Online Products | R@1: 84.46 -> 84.74 | Improved |
| Add Epipolar Loss to RRT + R50 | RRT + R50 trained on CO3D-Retrieve · SOP zero-shot | R@1: 76.32 -> 76.78 | Improved |
| Replace Epipolar Loss with Max-Epipolar Loss (ablation) | RRT + R50 · CO3D-Retrieve full images | R@1: 90.57 -> 90.69 | Improved |
| Replace Epipolar Loss with Max-Epipolar Loss (ablation) | RRT + R50 · CO3D-Retrieve masked backgrounds | R@1: 85.07 -> 85.17 | Improved |
| Replace Epipolar Loss with Max-Epipolar Loss (ablation) | RRT + R50 · Stanford Online Products | R@1: 84.74 -> 84.53 | Hurt |
| Use explicit Epipolar Positional Encoding instead of implicit Epipolar Loss (ablation) | RRT + R50 · CO3D-Retrieve full images | R@1: 90.57 -> 88.53 | Hurt |
| Use explicit Epipolar Positional Encoding instead of implicit Epipolar Loss (ablation) | RRT + R50 · Stanford Online Products | R@1: 84.74 -> 82.57 | Hurt |
| Increase local feature resolution from 7×7 to 14×14 (ablation) | RRT + R50 trained with LEPI · CO3D-Retrieve full images | R@1: 90.57 -> 90.71 | Improved |
Closest papers by meaning
- Understanding Multi-View Transformers (NeurIPS 2026 · similarity 0.87)
- GTA: A Geometry-Aware Attention Mechanism for Multi-View Transformers (ICLR 2024 · similarity 0.86)
- Cameras as Relative Positional Encoding (NeurIPS 2025 · similarity 0.85)
- Dissecting Query-Key Interaction in Vision Transformers (NeurIPS 2024 · similarity 0.84)
- Good Token Hunting: A Hitchhiker's Guide to Token Selection for Visual Geometry Transformers (NeurIPS 2026 · similarity 0.84)
- Multiple View Geometry Transformers for 3D Human Pose Estimation (CVPR 2024 · similarity 0.84)
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.
