aaaaaaaa Angle-I2P: Angle-Consistent-Aware Hierarchical Attention for Cross-Modality Outlier Rejection
Muyao Peng*, Shun Zou*, Pei An, You Yang, Qiong Liu · 2026
Read aaaaaaaa Angle-I2P: Angle-Consistent-Aware Hierarchical Attention for Cross-Modality Outlier Rejection by Muyao Peng*, Shun Zou*, Pei An, You Yang, Qiong Liu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-04541.txt.
Description
Peng and four colleagues teach machines to reject bad matches between images and point clouds. Image-to-point-cloud registration aligns a photo with a 3D scan, a core step for robot grasping and driving. Deep methods learn correspondences in a shared feature space, but when most initial matches are wrong, standard Perspective-n-Points solvers fail. Angle-I2P instead filters matches with a scale-invariant geometric constraint based on angular consistency, plus a hierarchical attention mechanism that works from global to local and drops geometrically inconsistent pairs under rigid motion.
The explicit angle constraint guides the network to separate inliers from outliers before pose estimation, lifting inlier ratio and registration recall. Experiments on the 7Scenes and RGBD Scenes V2 benchmarks and a self-collected dataset show top results with consistent gains across all three. For robots working from noisy data, cleaner matches mean more reliable localization.
Catalog record
| CREATOR | Muyao Peng*, Shun Zou*, Pei An, You Yang, Qiong Liu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.04541