$L^3$:Scene-agnostic Visual Localization in the Wild

TL;DR

This paper introduces $L^3$, a novel visual localization method that performs accurate scene localization in the wild without offline scene pre-processing by using online 3D reconstruction and pose refinement.

Contribution

$L^3$ is the first framework to achieve scene-agnostic localization without offline pre-processing, leveraging online 3D reconstruction and two-stage scale and pose refinement.

Findings

Performance comparable to state-of-the-art methods on benchmarks

Significantly more robust in sparse scenes with fewer reference images

Eliminates the need for offline scene representation storage

Abstract

Standard visual localization methods typically require offline pre-processing of scenes to obtain 3D structural information for better performance. This inevitably introduces additional computational and time costs, as well as the overhead of storing scene representations. Can we visually localize in a wild scene without any off-line preprocessing step? In this paper, we leverage the online inference capabilities of feed-forward 3D reconstruction networks to propose a novel map-free visual localization framework $L^3$. Specifically, by performing direct online 3D reconstruction on RGB images, followed by two-stage metric scale recovery and pose refinement based on 2D-3D correspondences, $L^3$ achieves high accuracy without the need to pre-build or store any offline scene representations. Extensive experiments demonstrate $L^3$ not only that the performance is comparable to state-of-the-art solutions on various benchmarks, but also that it exhibits significantly superior robustness in sparse scenes (fewer reference images per scene).