Multi-Sensor Corruption Benchmark (MSC-Bench)
收藏资源简介:
MSC-Bench是由北京人工智能研究院等机构创建的多传感器损坏基准测试数据集,旨在评估自动驾驶感知模型在传感器损坏情况下的鲁棒性。该数据集包含16种损坏类型,涵盖了天气、内部和传感器故障场景,基于nuScenes数据集的验证集构建。数据集通过模拟多种传感器损坏情况,如相机崩溃、帧丢失、交叉传感器干扰等,评估了六种3D目标检测模型和四种高精地图构建模型的性能。该数据集的应用领域主要集中在自动驾驶感知系统的鲁棒性评估,旨在解决传感器损坏对自动驾驶系统安全性的影响问题。
MSC-Bench is a multi-sensor corruption benchmark dataset developed by institutions including the Beijing Academy of Artificial Intelligence and other related institutions, aiming to evaluate the robustness of autonomous driving perception models under sensor corruption scenarios. This dataset includes 16 types of corruption cases covering weather-related issues, internal system faults and sensor failure scenarios, and is constructed based on the validation set of the nuScenes dataset. By simulating various sensor corruption situations such as camera crash, frame loss, cross-sensor interference and others, the dataset evaluates the performance of six 3D object detection models and four high-precision map construction models. The application scope of this dataset mainly focuses on the robustness evaluation of autonomous driving perception systems, with the goal of addressing the impact of sensor corruption on the safety of autonomous driving systems.




