AI Paper Atlas

Research map / Point clouds / Registration and matching

2D3D-MATR: 2D-3D Matching Transformer for Detection-Free Registration Between Images and Point Clouds

Minhao Li, Zheng Qin, Zhirui Gao, Renjiao Yi, Chenyang Zhu, Yulan Guo, Kai Xu

ICCV 2023 · Point clouds › Registration and matching · approach: Learned feature-based registration · 46 citations (OpenAlex)

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 changedWhereMeasuredEffect
add coarse-to-fine patch matching instead of matching random keypoints (ablation)2D3D-MATR · RGB-D Scenes V2Registration Recall: 34.6 -> 56.4Improved
add self-attention in feature refinement (ablation)2D3D-MATR · RGB-D Scenes V2Registration Recall: 44.0 -> 56.4Improved
add cross-attention in feature refinement (ablation)2D3D-MATR · RGB-D Scenes V2Registration Recall: 47.7 -> 56.4Improved
add attention modules overall (ablation)2D3D-MATR · RGB-D Scenes V2Inlier Ratio: 23.1 -> 32.4Improved
use 3-level multi-scale patch pyramid instead of a single resolution (ablation)2D3D-MATR · RGB-D Scenes V2Registration Recall: 36.9 -> 56.4 (vs 24x32 only)Improved
use the 6x8 coarse level in addition to finer levels (ablation)2D3D-MATR · RGB-D Scenes V2Patch Inlier Ratio: PIR 49.8 (12x16+6x8) vs 48.5 (full)Mixed
remove self-attention (ablation)2D3D-MATR · RGB-D Scenes V2Feature Matching Recall: 90.8 -> 91.8No clear effect
2D3D-MATR instead of P2-Net on RGB-D Scenes V2P2-Net · RGB-D Scenes V2Registration Recall: 38.4 -> 56.4Improved
2D3D-MATR instead of P2-Net on 7ScenesP2-Net · 7ScenesRegistration Recall: 65.7 -> 75.8Improved
predict overlap scores for keypoint sampling (Predator-2D3D) on 7ScenesPredator-2D3D · 7ScenesRegistration Recall: 61.4 -> 48.5 (vs FCGF-2D3D)Hurt

Later papers that compared with this method

8 later accepted papers report 2D3D-MATR as a baseline in their results tables; 8 of them report a better number on most of the results they share with it.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy