LiDAR-based 4D Occupancy Completion and Forecasting

The research introduces a novel LiDAR perception task of Occupancy Completion and Forecasting (OCF) in the field of autonomous driving. This task merges scene completion and forecasting into a unified framework. It highlights three challenges: sparse-to-dense reconstruction, partial-to-complete hallucination, and 3D-to-4D prediction. A new dataset called OCFBench has been created for this task. The research hopes to inspire further investigation into the rapidly evolving field of 4D perception.

Publication date: 18 Oct 2023
Project Page: https://github.com/ai4ce/Occ4cast
Paper: https://arxiv.org/pdf/2310.11239