[论文] GeoMix: Descriptor-Free Visual Localization via Global Context and Mul...
研究领域: CV 作者: Yejun Zhang, Xinjue Wang, Zihan Wang 发布时间: 2026-07-04 arXiv: 2507.03228
论文概要
研究领域: CV 作者: Yejun Zhang, Xinjue Wang, Zihan Wang 发布时间: 2026-07-04 arXiv: 2507.03228
中文摘要
无描述符视觉定位（Descriptor-free visual localization）消除了高维描述符存储、保护了场景隐私并简化了地图维护，但其精度仍远落后于基于描述符的管线。本文识别出这一差距的根源在于：纯几何匹配的几何可区分性（geometric discriminability）不足。由于缺乏视觉外观信息，现有方法未能充分利用局部几何线索，缺乏关键点之间的全局上下文，并过度拟合于单一关键点检测器。
本文进一步观察到，无描述符匹配天然支持多检测器训练（multi-detector training），因为异构关键点可在共享的纯几何空间中优化，无需对齐描述符空间。基于这些洞察，本文提出 GeoMix，一种在三个层次上增强几何可区分性的无描述符 2D-3D 匹配框架：
- 局部：方向感知和距离感知嵌入以细粒度空间结构丰富邻域聚合；
- 全局：可学习上下文节点通过交叉注意力聚合并重新分配场景级信息，解决超出局部感受野的歧义；
- 训练层面：Mix-Training 利用检测器无关的几何空间，在多个关键点检测器上学习表示。
原文摘要
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. We identify this gap to insufficient geometric discriminability in geometry-only matching. Without visual appearance, current methods underutilize local geometry cues, lack the global context among keypoints, and overfit to a single keypoint detector. We further observe that descriptor-free matching naturally enables multi-detector training, as heterogeneous keypoints can be optimized in a shared geometry-only space without aligning descriptor spaces. Building on these insights, we propose GeoMix, a descriptor-free 2D-3D matching framework that strengthens geometric discriminability at three levels. Locally, directional and distance-aware embeddings enrich neighborhood aggregation with fine-grained spatial structure. Globally, learnable context nodes aggregate and redistribute scen...
*自动采集于 2026-07-05*
#论文 #arXiv #CV #小凯