GeoMix: Descriptor-Free Visual Localization via Global Context and Multi-Detector Training

Forum topic · 小凯 · 2026-07-05

Summary

GeoMix (arXiv:2507.03228) is a descriptor-free 2D-3D matching framework for visual localization that closes the accuracy gap with descriptor-based pipelines. The authors attribute the performance gap of descriptor-free methods to insufficient geometric discriminability: without visual appearance, existing approaches underutilize local geometric cues, lack global context among keypoints, and overfit to a single keypoint detector. GeoMix addresses this at three levels: (1) locally, directional and distance-aware embeddings enrich neighborhood aggregation with fine-grained spatial structure; (2) globally, learnable context nodes aggregate and redistribute scene-level information via cross-attention to resolve ambiguities beyond local receptive fields; (3) at the training level, Mix-Training exploits a detector-agnostic geometry-only space to learn representations across multiple keypoint detectors, since heterogeneous keypoints can be optimized without aligning descriptor spaces. Experiments on MegaDepth, Cambridge Landmarks, 7Scenes, and Aachen Day-Night show state-of-the-art results among descriptor-free methods, reducing 75th-percentile rotation error by 89% and translation error by up to 90%, with zero-shot generalization to unseen detectors. The approach eliminates descriptor storage, preserves scene privacy, and simplifies map maintenance.

Paper Overview

Field: Computer Vision Authors: Yejun Zhang, Xinjue Wang, Zihan Wang Published: 2026-07-04 arXiv: 2507.03228

Abstract

Descriptor-free visual localization eliminates high-dimensional descriptor storage, preserves scene privacy, and simplifies map maintenance, yet its accuracy still lags far behind descriptor-based pipelines. The paper identifies this gap as stemming from insufficient geometric discriminability in geometry-only matching. Without visual appearance, current methods underutilize local geometric cues, lack global context among keypoints, and overfit to a single keypoint detector.

The authors further observe that descriptor-free matching naturally enables multi-detector training, since heterogeneous keypoints can be optimized in a shared geometry-only space without aligning descriptor spaces.

The GeoMix Framework

GeoMix is a descriptor-free 2D-3D matching framework that strengthens geometric discriminability at three levels:

Local level: directional and distance-aware embeddings enrich neighborhood aggregation with fine-grained spatial structure;
Global level: learnable context nodes aggregate and redistribute scene-level information via cross-attention, resolving ambiguities beyond local receptive fields;
Training level: Mix-Training leverages a detector-agnostic geometric space to learn representations across multiple keypoint detectors.

Results

Extensive experiments on MegaDepth, Cambridge Landmarks, 7Scenes, and Aachen Day-Night show that GeoMix achieves new state-of-the-art results among descriptor-free methods:

75th-percentile rotation error reduced by 89%;
Translation error reduced by up to 90%;
Zero-shot generalization to unseen detectors;
Significantly narrowed gap with descriptor-based pipelines.