Reproduce “WARM-3D: A Weakly-Supervised Sim2Real Domain Adaptation Framework for Roadside Monocular 3D Object Detection” 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
Xingcheng Zhou, Deyu Fu, Walter Zimmer, Mingyu Liu, Venkatnarayanan Lakshminarasimhan, Leah Strand, Alois C. Knoll
WARM-3D adapts monocular 3D detection from synthetic images to real images, using target 2D boxes to refine teacher predictions instead of target 3D boxes. It assumes calibration is available. On TUMTraf Intersection, mean average precision for 3D detection at 0.1 IoU is 41.55 versus 29.15 for source-only. With target 2D ground-truth boxes, it reports 48.09 versus 50.57 for a target-3D Oracle.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.