Research map / Point clouds / Registration and matching
UniCorrn: Unified Correspondence Transformer Across 2D and 3D
Prajnan Goswami, Tianye Ding, Feng Liu, Huaizu Jiang
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 |
|---|---|---|---|
| replace vanilla attention with Gaussian attention (ablation) | UniCorrn small model · MegaDepth-1500 (small-scale) | AUC@5: 36.4 -> 37.3 | Improved |
| increase keypoint queries from 100 to 800 (ablation) | UniCorrn small model · MegaDepth-1500 (small-scale) | AUC@5: 37.3 -> 38.2 | Improved |
| use single-head attention (H=1) (ablation) | UniCorrn small model · MegaDepth-1500 (small-scale) | AUC@5: 38.2 -> 39.5 | Improved |
| add contrastive loss (ablation) | UniCorrn small model · MegaDepth-1500 (small-scale) | AUC@5: 39.5 -> 43.9 | Improved |
| add 4x feature upsampling (ablation) | UniCorrn small model · MegaDepth-1500 (small-scale) | AUC@5: 43.9 -> 48.5 | Improved |
| add auxiliary supervision on intermediate predictions (1 decoder layer) (ablation) | UniCorrn small model · MegaDepth-1500 | AUC@5: 28.8 -> 47.7 | Improved |
| add auxiliary supervision on intermediate predictions (5 layers) (ablation) | UniCorrn small model · MegaDepth-1500 | AUC@5: 48.5 -> 50.6 | Improved |
| add pseudo point cloud training data (ablation) | UniCorrn · 7Scenes (2D-3D) | registration recall: 15.4 -> 77.8 | Improved |
| add pseudo point cloud training data (ablation) | UniCorrn · 3DLoMatch (3D-3D) | registration recall: 73.2 -> 81.8 | Improved |
| replace regression with the dual-stream attention matching decoder (ablation) | UniCorrn small model · MegaDepth (2D-2D) | AUC@5: 0.2 -> 50.6 | Improved |
| replace global matching with the proposed decoder (ablation) | UniCorrn small model · 7Scenes (2D-3D) | registration recall: 75.8 -> 77.8 | Improved |
| train jointly on all tasks rather than a single task (ablation) | UniCorrn · 7Scenes (2D-3D) | registration recall: 67.7 -> 91.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- 2D3D-MATR — 2D3D-MATR: 2D-3D Matching Transformer for Detection-Free Registration Between Images and Point Clouds (ICCV 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CorrNet3D: Unsupervised End-to-End Learning of Dense Correspondence for 3D Point Clouds (CVPR 2021 · similarity 0.88)
- UFM: A Simple Path towards Unified Dense Correspondence with Flow (NeurIPS 2025 · similarity 0.88)
- Tango3D: Towards Alignment for Global and Local 2D-3D Correspondence (NeurIPS 2026 · similarity 0.87)
- DV-Matcher: Deformation-based Non-rigid Point Cloud Matching Guided by Pre-trained Visual Features (CVPR 2025 · similarity 0.87)
- SemAlign3D: Semantic Correspondence between RGB-Images through Aligning 3D Object-Class Representations (CVPR 2025 · similarity 0.86)
- ODIN: A Single Model for 2D and 3D Segmentation (CVPR 2024 · similarity 0.86)
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.
