Research papers · 2025.06
Toward Real-world BEV Perception: Depth Uncertainty Estimation via Gaussian Splatting
Toward Real-world BEV Perception: Depth Uncertainty Estimation via Gaussian Splatting
Overview
GaussianLSS represents depth distributions as 3D Gaussians to construct uncertainty-aware BEV features. nuScenes experiments compare perception performance, execution speed and memory use with alternative projection and unprojection approaches.
Author-supplied source information
Author-supplied information on arXiv.
Accepted to CVPR'25. https://hcis-lab.github.io/GaussianLSS/
Version: arXiv v2 — manuscript revision