Leveraging Cycle-Consistent Anchor Points for Self-Supervised RGB-D Registration

2025/10/16 by Tourani, Siddharth, Reddy, Jayaram, Thakur, Sarvesh +3
#Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Robotics (cs.RO)

paper · doi:10.48550/arxiv.2510.14354

Abstract

With the rise in consumer depth cameras, a wealth of unlabeled RGB-D data has become available. This prompts the question of how to utilize this data for geometric reasoning of scenes. While many RGB-D registration meth- ods rely on geometric and feature-based similarity, we take a different approach. We use cycle-consistent keypoints as salient points to enforce spatial coherence constraints during matching, improving correspondence accuracy. Additionally, we introduce a novel pose block that combines a GRU recurrent unit with transformation synchronization, blending historical and multi-view data. Our approach surpasses previous self- supervised registration methods on ScanNet and 3DMatch, even outperforming some older supervised methods. We also integrate our components into existing methods, showing their effectiveness.

Citations

• Unsupervised Deep Probabilistic Approach for Partial Point Cloud Registration
• PCR-CG: Point Cloud Registration via Deep Explicit Color and Geometry
• Self-Supervised Correspondence Estimation via Multiview Registration
• Improving RGB-D Point Cloud Registration by Learning Multi-scale Local Linear Transformation
• Theseus: A Library for Differentiable Nonlinear Optimization
• Reliable Inlier Evaluation for Unsupervised Point Cloud Registration
• Geometric Transformer for Fast and Robust Point Cloud Registration
• Learning Pixel Trajectories with Multiscale Contrastive Random Walks
• DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Cameras
• Rethinking and Improving Relative Position Encoding for Vision Transformer
• Bootstrap Your Own Correspondences
• LoFTR: Detector-Free Local Feature Matching with Transformers
• PointDSC: Robust Point Cloud Registration using Deep Spatial Consistency
• UnsupervisedR&R: Unsupervised Point Cloud Registration via Differentiable Rendering
• PREDATOR: Registration of 3D Point Clouds with Low Overlap
• Unsupervised Point Cloud Registration via Salient Points Analysis (SPA)
• DeepGMR: Learning Latent Gaussian Mixture Models for Registration
• Space-Time Correspondence as a Contrastive Random Walk
• Feature-metric Registration: A Fast Semi-supervised Approach for Robust Point Cloud Registration without Correspondences
• Learning Feature Descriptors using Camera Pose Supervision
• Deep Global Registration
• PointGMM: a Neural GMM Network for Point Clouds
• D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features
• Learning multiview 3D point cloud registration
• SuperGlue: Learning Feature Matching with Graph Neural Networks
• Deep Closest Point: Learning Representations for Point Cloud Registration
• SDRSAC: Semidefinite-Based Randomized Approach for Robust Point Cloud Registration without Correspondences
• On the Continuity of Rotation Representations in Neural Networks
• The Perfect Match: 3D Point Cloud Matching with Smoothed Densities
• PPF-FoldNet: Unsupervised Learning of Rotation Invariant 3D Local Descriptors
• Attention Is All You Need
• ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes
• 3DMatch: Learning Local Geometric Descriptors from RGB-D Reconstructions
• Deep Residual Learning for Image Recognition
• Adam: A Method for Stochastic Optimization
• Distinctive Image Features from Scale-Invariant Keypoints

Cited by

• Leveraging 2D Priors and SDF Guidance for Dynamic Urban Scene Rendering

Related

• ViGG: Robust RGB-D Point Cloud Registration using Visual-Geometric Mutual Guidance
• Geometry Depth Consistency in RGBD Relative Pose Estimation
• Improving RGB-D Point Cloud Registration by Learning Multi-scale Local Linear Transformation
• ObjectMatch: Robust Registration using Canonical Object Correspondences
• PointMBF: A Multi-scale Bidirectional Fusion Network for Unsupervised RGB-D Point Cloud Registration