ImLoc: Revisiting Visual Localization with Image-based Representation

2026-01-07

arXiv.org

AI Summary

This work addresses the longstanding trade-off in visual localization between 2D representations, which are lightweight and easy to maintain but lack strong geometric reasoning, and 3D representations, which offer high accuracy but are difficult to update and scale. To bridge this gap, the authors propose an enhanced 2D image representation that incorporates estimated depth maps to inject geometric structure while preserving the efficiency and scalability of 2D systems. By integrating dense feature matching, compact map compression, and a GPU-accelerated LO-RANSAC algorithm, the method achieves state-of-the-art accuracy across multiple standard benchmarks. It significantly outperforms existing memory-efficient approaches and establishes a new balance between precision and computational efficiency at comparable map sizes.

Technology Category

• Computer Vision: Representation Learning for Vision
• Intelligent Robots: Localization, Mapping, and Navigation
• Machine Learning: Representation Learning

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Semantics and Knowledge: Representation, semantic annotation, enhancement, enrichments, access and/or integration of a variety of data on the Web

Abstract

Existing visual localization methods are typically either 2D image-based, which are easy to build and maintain but limited in effective geometric reasoning, or 3D structure-based, which achieve high accuracy but require a centralized reconstruction and are difficult to update. In this work, we revisit visual localization with a 2D image-based representation and propose to augment each image with estimated depth maps to capture the geometric structure. Supported by the effective use of dense matchers, this representation is not only easy to build and maintain, but achieves highest accuracy in challenging conditions. With compact compression and a GPU-accelerated LO-RANSAC implementation, the whole pipeline is efficient in both storage and computation and allows for a flexible trade-off between accuracy and highest memory efficiency. Our method achieves a new state-of-the-art accuracy on various standard benchmarks and outperforms existing memory-efficient methods at comparable map sizes. Code will be available at https://github.com/cvg/Hierarchical-Localization.

Problem

• visual localization
• image-based representation
• geometric reasoning
• 3D reconstruction
• memory efficiency

Innovation

• image-based localization
• depth-augmented representation
• dense matching
• LO-RANSAC
• memory-efficient mapping