DIO: Decomposable Implicit 4D Occupancy-Flow World Model

Christopher Diehl; Quinlan Sykora; Ben Agro; Thomas Gilles; Sergio Casas; Raquel Urtasun

doi:10.1109/cvpr52734.2025.02557

CVPR2025顶会

DIO: Decomposable Implicit 4D Occupancy-Flow World Model

Christopher Diehl, Quinlan Sykora, Ben Agro, Thomas Gilles, Sergio Casas, Raquel Urtasun

DOI 出版方

2025年份

摘要

Scene Occupancy (b) Instance Occupancy (c) Scene Flow (d) Rendered LiDAR Figure 1. We present DIO, a world model that learns unsupervised 4D occupancy and flow (a, c) and can be decomposed into instance occupancy (b) as well as transfered to downstream tasks like LiDAR point cloud forecasting (d).