Abstract
We introduce UniOcc, a comprehensive, unified benchmark and toolkit foroccupancy forecasting (i.e., predicting future occupancies based on historicalinformation) and occupancy prediction (i.e., predicting current-frame occupancyfrom camera images. UniOcc unifies the data from multiple real-world datasets(i.e., nuScenes, Waymo) and high-fidelity driving simulators (i.e., CARLA,OpenCOOD), providing 2D/3D occupancy labels and annotating innovative per-voxelflows. Unlike existing studies that rely on suboptimal pseudo labels forevaluation, UniOcc incorporates novel evaluation metrics that do not depend onground-truth labels, enabling robust assessment on additional aspects ofoccupancy quality. Through extensive experiments on state-of-the-art models, wedemonstrate that large-scale, diverse training data and explicit flowinformation significantly enhance occupancy prediction and forecastingperformance. Our data and code are available at https://uniocc.github.io/.