Learning-Deep-Learning

Occ3D: A Large-Scale 3D Occupancy Prediction Benchmark for Autonomous Driving

May 2023

tl;dr: Creation of two semantic occupancy prediction datasets, Occ3D-Waymo and Occ3D-nuScenes.

Overall impression

Occ3D also proposed a pipeline to generate dense occupancy labels, which includes point cloud aggregation, point labeling, and occlusion handling. The visibility and occlusion reasoning of the label is the main contribution of the paper.

It does not have the densification process in SurroundOcc and OpenOccupancy, which focused on NuScenes dataset. The authors claim that the label is already quite dense even without densification for Waymo dataset, and Poisson Recon leads to inaccurate annotation.

The paper also proposed a neural network architecture Coarse-to-Fine Occupancy (CTF-Occ). This is largely the same as Cascade Occupancy Net (OCNet) by OpenOccupancy and the coarse to fine architecture of SurroundOcc. It proposes two tricks: incremental token selection to reduce computation burden, and an implicit decoder to output the semantic label of any given point, similar to the idea of Occupancy Networks.

Key ideas

Technical details

Notes