AsymLoc: Towards Asymmetric Feature Matching for Efficient Visual Localization

TL;DR

AsymLoc introduces an asymmetric feature matching framework for efficient visual localization, enabling lightweight models to achieve high accuracy comparable to larger models through distillation techniques.

Contribution

The paper presents a novel distillation framework that aligns a lightweight Student model with a large Teacher model for fast, accurate visual localization on resource-constrained devices.

Findings

Achieves up to 95% of teacher accuracy with much smaller models.

Significantly outperforms existing baselines in efficiency-accuracy trade-off.

Establishes new state-of-the-art performance on multiple benchmarks.

Abstract

Precise and real-time visual localization is critical for applications like AR/VR and robotics, especially on resource-constrained edge devices such as smart glasses, where battery life and heat dissipation can be a primary concerns. While many efficient models exist, further reducing compute without sacrificing accuracy is essential for practical deployment. To address this, we propose asymmetric visual localization: a large Teacher model processes pre-mapped database images offline, while a lightweight Student model processes the query image online. This creates a challenge in matching features from two different models without resorting to heavy, learned matchers. We introduce AsymLoc, a novel distillation framework that aligns a Student to its Teacher through a combination of a geometry-driven matching objective and a joint detector-descriptor distillation objective, enabling fast, parameter-less nearest-neighbor matching. Extensive experiments on HPatches, ScanNet, IMC2022, and Aachen show that AsymLoc achieves up to 95% of the teacher's localization accuracy using an order of magnitude smaller models, significantly outperforming existing baselines and establishing a new state-of-the-art efficiency-accuracy trade-off.