Scene Coordinate Reconstruction: Posing of Image Collections via Incremental Learning of a Relocalizer

TL;DR

This paper introduces ACE0, a learning-based relocalization method that reconstructs scenes and estimates camera poses from unposed images without requiring pose priors, achieving accuracy comparable to traditional SfM methods.

Contribution

It presents a novel scene coordinate regression approach that builds implicit neural scene representations from unposed images, bypassing the need for local feature matching or pose priors.

Findings

ACE0 estimates camera poses with accuracy close to feature-based SfM.

The method efficiently handles thousands of images without pose priors.

It enables novel view synthesis demonstrating high-quality scene reconstruction.

Abstract

We address the task of estimating camera parameters from a set of images depicting a scene. Popular feature-based structure-from-motion (SfM) tools solve this task by incremental reconstruction: they repeat triangulation of sparse 3D points and registration of more camera views to the sparse point cloud. We re-interpret incremental structure-from-motion as an iterated application and refinement of a visual relocalizer, that is, of a method that registers new views to the current state of the reconstruction. This perspective allows us to investigate alternative visual relocalizers that are not rooted in local feature matching. We show that scene coordinate regression, a learning-based relocalization approach, allows us to build implicit, neural scene representations from unposed images. Different from other learning-based reconstruction methods, we do not require pose priors nor sequential inputs, and we optimize efficiently over thousands of images. In many cases, our method, ACE0, estimates camera poses with an accuracy close to feature-based SfM, as demonstrated by novel view synthesis. Project page: https://nianticlabs.github.io/acezero/