Decision PCR: Decision version of the Point Cloud Registration task

2025-07-20

AI Summary

In low-overlap point cloud registration (PCR), conventional inlier-count-based evaluation metrics fail under extremely low overlap ratios. This paper introduces, for the first time, a decision-oriented formulation of PCR—framing registration quality assessment as a binary classification task (success vs. failure)—and constructs the first dedicated benchmark dataset for this task, built upon 3DMatch with systematic low-overlap augmentation. We propose an end-to-end deep learning classifier, designed to be seamlessly embedded into arbitrary registration pipelines for automatic, threshold-free quality assessment. Crucially, our approach eliminates reliance on geometric priors and hand-crafted thresholds. Evaluated on the 3DLoMatch benchmark, it achieves a state-of-the-art registration recall of 86.97%. Furthermore, strong generalization is demonstrated on the ETH outdoor dataset, confirming robustness across diverse real-world scenarios.

Technology Category

• Computer Vision: 3D Computer Vision
• Machine Learning: Calibration & Uncertainty Quantification
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Security and Privacy: Large-scale security measurements

Abstract

Low-overlap point cloud registration (PCR) remains a significant challenge in 3D vision. Traditional evaluation metrics, such as Maximum Inlier Count, become ineffective under extremely low inlier ratios. In this paper, we revisit the registration result evaluation problem and identify the Decision version of the PCR task as the fundamental problem. To address this Decision PCR task, we propose a data-driven approach. First, we construct a corresponding dataset based on the 3DMatch dataset. Then, a deep learning-based classifier is trained to reliably assess registration quality, overcoming the limitations of traditional metrics. To our knowledge, this is the first comprehensive study to address this task through a deep learning framework. We incorporate this classifier into standard PCR pipelines. When integrated with our approach, existing state-of-the-art PCR methods exhibit significantly enhanced registration performance. For example, combining our framework with GeoTransformer achieves a new SOTA registration recall of 86.97% on the challenging 3DLoMatch benchmark. Our method also demonstrates strong generalization capabilities on the unseen outdoor ETH dataset.

Problem

• Addresses low-overlap point cloud registration challenges
• Proposes data-driven evaluation for registration quality
• Enhances existing PCR methods with deep learning

Innovation

• Data-driven approach for Decision PCR task
• Deep learning classifier for registration quality
• Integration with standard PCR pipelines