Reproduce “VESPA: Towards un(Human)supervised Open-World Pointcloud Labeling for Autonomous Driving” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Levente Tempfli, Esteban Rivera, Markus Lienkamp
VESPA uses image models to identify and segment objects, projects camera masks onto LiDAR points, and corrects mismatches through clustering and temporal refinement to generate 3D bounding-box labels. It reports 52.95 AP and 48.12 NDS versus UNION’s 38.4 AP and 31.2 NDS for class-agnostic nuScenes detection. These validation results use pseudo-label-trained detectors and do not validate novel categories.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.