OmniHD-Scenes
收藏资源简介:
该数据集名为OmniHD-Scenes,包含了1.5千个序列,这些序列是在多种驾驶场景下,由六个多视角摄像头和4D雷达捕捉的。其中,200个序列被标注,总计11921个关键帧,带有全面的占用地面真实标签,这为评估多模态占用预测提供了可能。此外,该数据集还用于通过半监督学习方法构建了OmniHD-SemiOcc数据集。其规模包括1.5千个序列、200个标注序列以及11921个关键帧。该数据集的任务是进行3D占用预测。
This dataset, named OmniHD-Scenes, contains 1.5 thousand sequences captured by six multi-view cameras and a 4D radar across diverse driving scenarios. 200 of these sequences are annotated, with a total of 11,921 key frames equipped with comprehensive occupancy ground-truth labels, which enables the evaluation of multimodal occupancy prediction. Additionally, this dataset is utilized to construct the OmniHD-SemiOcc dataset via semi-supervised learning methods. Its scale encompasses 1.5 thousand sequences, 200 annotated sequences, and 11,921 key frames. The core task of this dataset is 3D occupancy prediction.




