DCI数据集
收藏arXiv2023-04-11 更新2024-08-06 收录
下载链接：
http://arxiv.org/abs/2304.05098v1
下载链接
链接失效反馈官方服务：
资源简介：
DCI数据集是由北京航空航天大学创建，用于评估物理世界中车辆检测模型的对抗性鲁棒性。该数据集包含7个连续场景和1个离散场景，涵盖超过40个角度、20个距离和20,000个位置，采用多视角和多天气条件进行数据采集。创建过程中，利用CARLA模拟器结合神经渲染技术生成高保真场景。DCI数据集主要应用于自动驾驶领域，旨在解决检测模型在面对物理世界对抗攻击时的鲁棒性问题。
The DCI Dataset was developed by Beihang University to evaluate the adversarial robustness of vehicle detection models in the physical world. This dataset contains 7 continuous scenarios and 1 discrete scenario, covering over 40 angles, 20 distances, and 20,000 positions, with data collected under multi-view and diverse weather conditions. During the creation process, the CARLA simulator combined with neural rendering techniques was employed to generate high-fidelity scenarios. The DCI Dataset is primarily applied in the field of autonomous driving, aiming to address the robustness issues of detection models when facing adversarial attacks in the physical world.
提供机构：
北京航空航天大学计算机科学与工程学院
创建时间：
2023-04-11
