Shu-Wei Lu
Research Scientist @ National Yang Ming Chiao Tung University
Research Scientist @ National Yang Ming Chiao Tung University
I’m a research scientist working on 3D perception and scene reconstruction for autonomous driving. I’m interested in bridging the gap between visual understanding and downstream planning via 3D representations, simulation, and neural rendering.
GaussianLSS - Toward Real-world BEV Perception: Depth Uncertainty Estimation via Gaussian Splatting
CVPR, 2025
project page / paper / code
We propose GaussianLSS, an uncertainty-aware BEV perception framework that revisits the Lift-Splat-Shoot paradigm and enhances it with depth distribution modeling. By transforming per-pixel depth distributions into 3D Gaussians, our method constructs spatially-aware BEV features that are both efficient and accurate.
Potential Field as Scene Affordance for Behavior Change-Based Visual Risk Object Identification
ICRA, 2025
project page / paper / code
We study behavior change-based visual risk object identification (Visual-ROI), a critical framework designed to detect potential hazards for intelligent driving systems.
Action-slot: Visual Action-centric Representations for Atomic Activity Recognition in Traffic Scenes
CVPR, 2024
project page / CVF / arXiv / code / TACO dataset
We use Action-slot to represent atomic activities. The learned attention can discover and localize atomic activities with only weak video labels and without using any perception module (e.g., object detector).
RiskBench: A Scenario-based Benchmark for Risk Identification
ICRA, 2024
project page / paper / code / dataset
The FIRST benchmark that enables evaluation of various types of risk identification algorithms, including rule-based, trajectory-prediction-based, collision prediction, and behavior-change-based methods. We also assess the influence of risk identification on downstream driving tasks.