ImLoc: Revisiting Visual Localization with Image-based Representation

TL;DR

ImLoc reframes visual localization by keeping a 2D image-based map augmented with dense depth maps, enabling strong geometric reasoning without building a global 3D model. The pipeline uses retrieval for coarse localization, dense per-pixel matching with RoMa to generate dense 2D-2D correspondences, which are lifted to 2D-3D via precomputed depth maps and solved with a GPU-accelerated LO-RANSAC PnP, yielding state-of-the-art accuracy under memory-efficient map sizes. The approach showcases robust performance across diverse benchmarks (Oxford Day & Night, LaMAR, Cambridge, Aachen) and supports flexible memory-accuracy trade-offs (nano/micro maps) through aggressive compression and configurable depth/rgb resolutions. The work emphasizes simplicity, adaptability, and scalability, offering a practical baseline for robust, storage-efficient visual localization in dynamic environments and providing code and models for reproducibility. Overall, ImLoc demonstrates that depth-augmented 2D image representations coupled with dense matching can rival or surpass more complex 3D or NVS-based methods while maintaining easy maintenance and updateability.

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.

Figures (14)

• Figure 1: Left:ImLoc achieves state-of-the-art results on a multitude of datasets: LaMAR sarlin2022lamar, Aachen Day and Night 1.1 sattler2012bmvcsattler2018cvpr, Oxford Day and Night wang2025seeing, Cambridge Landmarks kendall2015posenet, surpassing the previous gold standard HLoc sarlin2019coarse. Right:ImLoc (${\color{Maroon} \star}, {\color{Salmon} \star}, {\color{Red} \star}$) allows a trade-off between accuracy and memory efficiency and maintains state-of-the-art accuracy at various compression levels.
• Figure 2: Illustration of our localization pipeline. During mapping, we store RGB and depth images along with camera poses, intrinsics, and retrieval features. For localization, we run dense image matching between the query and the top-K retrieved database images and establish 2D-3D correspondences using the precomputed depth maps. The camera pose is estimated with PnP+RANSAC. Please refer to Section \ref{['sec:pipeline']} for details.
• Figure 3: RGB Image Subsampling and Compression on Cambridge. Localization can tolerate low quality setting for modern image compression like JPEG XL but it is more sensitive to keyframe subsampling or downsampling resolution, which decreases performance.
• Figure 4: Depth Image Compression on Cambridge. The quantization usually saturates at 8bit quantization (256 depth levels), while a lower depth resolution does not significantly affect performance for any of the selected quantization levels.
• Figure 5: Performance of different retrieval methods and number of retrieved images on LaMAR sarlin2022lamar. We plot the percentage of poses with error smaller than 1m and 5$^{\circ}$.