AsymLoc: Towards Asymmetric Feature Matching for Efficient Visual Localization
2026-04-10 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors address the challenge of making visual localization work well on small devices like smart glasses, which have limited battery and compute power. They propose a method called AsymLoc where a big, powerful model processes the known images ahead of time, while a small, fast model handles new images in real-time. Their method teaches the small model to match the big model's results using clever techniques that don't require heavy computations. Experiments show their approach keeps most of the big model's accuracy but uses much less computing power, doing better than previous methods.
visual localizationdistillation frameworkteacher-student modelfeature matchinggeometry-driven matchingdetector-descriptornearest-neighbor matchingedge devicesmodel efficiencyAR/VR
Authors
Mohammad Omama, Gabriele Berton, Eric Foxlin, Yelin Kim
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.