OCCUQ
收藏资源简介:
OCCUQ数据集是由亚琛工业大学汽车工程研究所创建的,用于探索高效不确定性量化的3D占用预测研究。该数据集包含了6019个场景,使用6个摄像头捕获的多视角图像和语义分割的激光雷达点云作为输入,为3D占用预测任务提供数据支持。数据集通过模拟不同的摄像头缺陷来评估模型在未知数据上不确定性量化的能力,并用于验证高效不确定性估计方法在现实世界场景下的鲁棒性。
The OCCUQ dataset was created by the Institute of Automotive Engineering of RWTH Aachen University for research on efficient uncertainty quantification for 3D occupancy prediction. It comprises 6019 scenes, taking multi-view images captured by six cameras and semantically segmented LiDAR point clouds as inputs to provide data support for the 3D occupancy prediction task. The dataset evaluates the model's uncertainty quantification performance on out-of-distribution data by simulating various camera defects, and is employed to validate the robustness of efficient uncertainty estimation methods in real-world scenarios.

- 1OCCUQ: Exploring Efficient Uncertainty Quantification for 3D Occupancy Prediction亚琛工业大学汽车工程研究所 · 2025年



