Toward Real-world BEV Perception: Depth Uncertainty Estimation via Gaussian Splatting
Shu-Wei Lu, Yi-Hsuan Tsai, Yi-Ting Chen · 2025
Read Toward Real-world BEV Perception: Depth Uncertainty Estimation via Gaussian Splatting by Shu-Wei Lu, Yi-Hsuan Tsai, Yi-Ting Chen (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-2504-01957.txt.
Description
Shu-Wei Lu and co-authors build GaussianLSS, a bird's-eye view perception system for self-driving cars. It fuses camera views into one top-down map. Newer systems learn the map with queries but skip uncertainty and cost too much to run. This one returns to an older lift-and-splat design and adds depth uncertainty.
Each pixel's depth becomes a soft estimate with a spread, not a single guess. The spread captures how far an object extends. Those estimates turn into 3D Gaussians, then splat into the map. On the nuScenes benchmark the system led its class. It ran 2.5 times faster than query-based rivals and used a fraction of their memory, at near-equal accuracy.
Catalog record
| CREATOR | Shu-Wei Lu, Yi-Hsuan Tsai, Yi-Ting Chen |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-01957