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

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tourani, Siddharth, Reddy, Jayaram, Thakur, Sarvesh, Krishna, K Madhava, Khan, Muhammad Haris, Reddy, N Dinesh
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914096257957888
author Tourani, Siddharth
Reddy, Jayaram
Thakur, Sarvesh
Krishna, K Madhava
Khan, Muhammad Haris
Reddy, N Dinesh
author_facet Tourani, Siddharth
Reddy, Jayaram
Thakur, Sarvesh
Krishna, K Madhava
Khan, Muhammad Haris
Reddy, N Dinesh
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2510_14354
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Leveraging Cycle-Consistent Anchor Points for Self-Supervised RGB-D Registration
Tourani, Siddharth
Reddy, Jayaram
Thakur, Sarvesh
Krishna, K Madhava
Khan, Muhammad Haris
Reddy, N Dinesh
Computer Vision and Pattern Recognition
Robotics
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.
title Leveraging Cycle-Consistent Anchor Points for Self-Supervised RGB-D Registration
topic Computer Vision and Pattern Recognition
Robotics
url https://arxiv.org/abs/2510.14354