UnsupervisedR&R: Unsupervised Point Cloud Registration via Differentiable Rendering

TL;DR

UnsupervisedR&R introduces an end-to-end unsupervised method for point cloud registration from RGB-D videos, leveraging differentiable rendering to enforce consistency, outperforming traditional methods and competing with supervised approaches.

Contribution

The paper presents a novel unsupervised learning framework for point cloud registration that does not require pose annotations, using differentiable rendering for consistency enforcement.

Findings

Outperforms traditional registration methods on indoor datasets

Competitive with supervised geometric registration approaches

Effective in leveraging raw RGB-D data without annotations

Abstract

Aligning partial views of a scene into a single whole is essential to understanding one's environment and is a key component of numerous robotics tasks such as SLAM and SfM. Recent approaches have proposed end-to-end systems that can outperform traditional methods by leveraging pose supervision. However, with the rising prevalence of cameras with depth sensors, we can expect a new stream of raw RGB-D data without the annotations needed for supervision. We propose UnsupervisedR&R: an end-to-end unsupervised approach to learning point cloud registration from raw RGB-D video. The key idea is to leverage differentiable alignment and rendering to enforce photometric and geometric consistency between frames. We evaluate our approach on indoor scene datasets and find that we outperform existing traditional approaches with classic and learned descriptors while being competitive with supervised geometric point cloud registration approaches.