Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PSReg: Prior-guided Sparse Mixture of Experts for Point Cloud Registration

Xiaoshui Huang, Zhou Huang, Yifan Zuo, Yongshun Gong, Chengdong Zhang, Deyang Liu, Yuming Fang

TL;DR

This work tackles partial-overlap point cloud registration by introducing a Prior-guided Sparse Mixture of Experts (PSMoE) that routes tokens to specialized experts using priors on overlap and potential correspondences. Integrated into a coarse-to-fine PSReg framework with a Transformer backbone, the method employs a Prior Superpoint Correspondence Encoding (PCE) to encode overlap priors and guide routing, improving feature distinctiveness in overlapping regions. Experiments on 3DMatch/3DLoMatch and ModelNet(ModelLoNet) demonstrate state-of-the-art registration recall and robust inlier ratios, with ablations confirming the effectiveness of ordered coding and prior-guided routing. Overall, the paper shows SMoE can enhance 3D registration by targeted, prior-informed routing that mitigates ambiguity in overlaps, enabling more reliable correspondences and transformations.

Abstract

The discriminative feature is crucial for point cloud registration. Recent methods improve the feature discriminative by distinguishing between non-overlapping and overlapping region points. However, they still face challenges in distinguishing the ambiguous structures in the overlapping regions. Therefore, the ambiguous features they extracted resulted in a significant number of outlier matches from overlapping regions. To solve this problem, we propose a prior-guided SMoE-based registration method to improve the feature distinctiveness by dispatching the potential correspondences to the same experts. Specifically, we propose a prior-guided SMoE module by fusing prior overlap and potential correspondence embeddings for routing, assigning tokens to the most suitable experts for processing. In addition, we propose a registration framework by a specific combination of Transformer layer and prior-guided SMoE module. The proposed method not only pays attention to the importance of locating the overlapping areas of point clouds, but also commits to finding more accurate correspondences in overlapping areas. Our extensive experiments demonstrate the effectiveness of our method, achieving state-of-the-art registration recall (95.7\%/79.3\%) on the 3DMatch/3DLoMatch benchmark. Moreover, we also test the performance on ModelNet40 and demonstrate excellent performance.

PSReg: Prior-guided Sparse Mixture of Experts for Point Cloud Registration

TL;DR

This work tackles partial-overlap point cloud registration by introducing a Prior-guided Sparse Mixture of Experts (PSMoE) that routes tokens to specialized experts using priors on overlap and potential correspondences. Integrated into a coarse-to-fine PSReg framework with a Transformer backbone, the method employs a Prior Superpoint Correspondence Encoding (PCE) to encode overlap priors and guide routing, improving feature distinctiveness in overlapping regions. Experiments on 3DMatch/3DLoMatch and ModelNet(ModelLoNet) demonstrate state-of-the-art registration recall and robust inlier ratios, with ablations confirming the effectiveness of ordered coding and prior-guided routing. Overall, the paper shows SMoE can enhance 3D registration by targeted, prior-informed routing that mitigates ambiguity in overlaps, enabling more reliable correspondences and transformations.

Abstract

The discriminative feature is crucial for point cloud registration. Recent methods improve the feature discriminative by distinguishing between non-overlapping and overlapping region points. However, they still face challenges in distinguishing the ambiguous structures in the overlapping regions. Therefore, the ambiguous features they extracted resulted in a significant number of outlier matches from overlapping regions. To solve this problem, we propose a prior-guided SMoE-based registration method to improve the feature distinctiveness by dispatching the potential correspondences to the same experts. Specifically, we propose a prior-guided SMoE module by fusing prior overlap and potential correspondence embeddings for routing, assigning tokens to the most suitable experts for processing. In addition, we propose a registration framework by a specific combination of Transformer layer and prior-guided SMoE module. The proposed method not only pays attention to the importance of locating the overlapping areas of point clouds, but also commits to finding more accurate correspondences in overlapping areas. Our extensive experiments demonstrate the effectiveness of our method, achieving state-of-the-art registration recall (95.7\%/79.3\%) on the 3DMatch/3DLoMatch benchmark. Moreover, we also test the performance on ModelNet40 and demonstrate excellent performance.
Paper Structure (24 sections, 7 equations, 7 figures, 4 tables)

This paper contains 24 sections, 7 equations, 7 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Non-Transformer-based Registration
  4. Transformer-based Registration
  5. Sparse Mixture of Experts (SMoE)
  6. Method
  7. Preliminary and Analysis
  8. Point Cloud Registration.
  9. Vanilla SMoE.
  10. Analyze Vanilla SMoE.
  11. PSMoE Module
  12. Prior Superpoint Correspondence Prediction.
  13. Prior Superpoint Correspondence Encoding (PCE).
  14. Prior-guided Routing.
  15. PSReg
  16. ...and 9 more sections

Figures (7)

  • Figure 1: Performance gaps in applying vanilla SMoE and our PSMoE. Fig. (a) and (c) show coarse patch correspondences, (b) and (d) show fine point correspondences.
  • Figure 2: The visualization of the routing of tokens (patches) in the last SMoE block, where the color of the point cloud patches indicates the selected experts.
  • Figure 3: Difference between vanilla SMoE and PSMoE.
  • Figure 4: Diagram of PCE module.
  • Figure 5: The framework of PSReg. The red blocks represent superpoints located in the putative non-overlapping areas.
  • ...and 2 more figures