BevSplat: Resolving Height Ambiguity via Feature-Based Gaussian Primitives for Weakly-Supervised Cross-View Localization
Qiwei Wang, Shaoxun Wu, Yujiao Shi · 2025
Read BevSplat: Resolving Height Ambiguity via Feature-Based Gaussian Primitives for Weakly-Supervised Cross-View Localization by Qiwei Wang, Shaoxun Wu, Yujiao Shi (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-09080.txt.
Description
Qiwei Wang, Shaoxun Wu, and Yujiao Shi wrote a 2025 computer vision paper on BevSplat, a method for locating a ground camera against a satellite image. Cross view localization matters for navigation where GPS fails. The hard part is height: a ground photo and a flat satellite view disagree about how tall things are, and depth cues are missing.
BevSplat represents each ground image pixel as a 3D Gaussian blob carrying meaning and position features. These blobs are splatted into a bird's eye view feature map, from which the camera pose is read. A supervision trick based on a sphere mesh handles wide panoramic queries. The method was validated on two standard datasets covering both narrow and panoramic cameras. It avoids assuming flat ground.
Catalog record
| CREATOR | Qiwei Wang, Shaoxun Wu, Yujiao Shi |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 83 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-09080