This website requires JavaScript.

SuperGF: Unifying Local and Global Features for Visual Localization

Wenzheng SongRan YanBoshu LeiTakayuki Okatani
Dec 2022
摘要
Advanced visual localization techniques encompass image retrieval challengesand 6 Degree-of-Freedom (DoF) camera pose estimation, such as hierarchicallocalization. Thus, they must extract global and local features from inputimages. Previous methods have achieved this through resource-intensive oraccuracy-reducing means, such as combinatorial pipelines or multi-taskdistillation. In this study, we present a novel method called SuperGF, whicheffectively unifies local and global features for visual localization, leadingto a higher trade-off between localization accuracy and computationalefficiency. Specifically, SuperGF is a transformer-based aggregation model thatoperates directly on image-matching-specific local features and generatesglobal features for retrieval. We conduct experimental evaluations of ourmethod in terms of both accuracy and efficiency, demonstrating its advantagesover other methods. We also provide implementations of SuperGF using varioustypes of local features, including dense and sparse learning-based orhand-crafted descriptors.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?