Reproduce “Diffusion-Based Generative Models for 3D Occupancy Prediction in 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
Yunshen Wang, Yicheng Liu, Tianyuan Yuan, Yingshi Liang, Xiuyu Yang, Honggang Zhang, Hang Zhao
DiffOcc changes camera-based 3D occupancy prediction by denoising a noisy categorical occupancy grid conditioned on multi-view image features. A BEV encoder runs once, while a U-Net denoiser refines the grid. On Occ3D-nuScenes validation, matched R101 BEVFormer* and DiffOcc score 35.62 and 42.67 mIoU; on labeled camera-invisible voxels, 21.42 and 36.45. LiDAR-derived labels may miss regions, limiting reconstruction claims.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.