KITTI-AR
收藏资源简介:
KITTI-AR数据集是由上海大学通信与信息工程学院的研究团队创建的,用于3D异常物体检测。该数据集基于现有的KITTI数据集,通过3D渲染方法增加了97个新的类别,总共包含6千对立体图像。此外,KITTI-AR-ExD子集包含了39个常见类别作为额外训练数据,而KITTI-AR-OoD子集则包含了58个稀有类别,用于模拟真实世界中的零样本场景。该数据集旨在验证和提升异常检测的泛化能力。
The KITTI-AR dataset was developed by the research team from the School of Communication and Information Engineering, Shanghai University, for 3D anomalous object detection. Built upon the established KITTI dataset, this dataset adds 97 new categories via 3D rendering techniques, with a total of 6,000 pairs of stereo images. In addition, the KITTI-AR-ExD subset provides 39 common categories as supplementary training data, while the KITTI-AR-OoD subset contains 58 rare categories designed to simulate real-world zero-shot scenarios. The primary goal of this dataset is to validate and enhance the generalization capability of anomaly detection systems.




