The paper introduces HoloVIC, a large-scale multi-sensor holographic intersection and vehicle-infrastructure cooperation dataset. The dataset is constructed by creating holographic intersections with various layouts, equipped with different types of sensors (Camera, Lidar, Fisheye) to capture synchronized data. Each intersection is equipped with 6-18 sensors, and the dataset includes over 100k synchronous frames from different sensors. The authors annotate 3D bounding boxes for each sensor and associate the same objects across different devices and frames. They also formulate four tasks—Monocular 3D Detection, Lidar 3D Detection, Multiple Object Tracking, and Multi-sensor Multi-object Tracking—to facilitate research in roadside perception and vehicle-infrastructure cooperation. The paper provides benchmarks for these tasks and evaluates the performance of various methods using the HoloVIC dataset. The results show that the proposed dataset and benchmarks can effectively evaluate and improve the performance of perception models in autonomous driving.The paper introduces HoloVIC, a large-scale multi-sensor holographic intersection and vehicle-infrastructure cooperation dataset. The dataset is constructed by creating holographic intersections with various layouts, equipped with different types of sensors (Camera, Lidar, Fisheye) to capture synchronized data. Each intersection is equipped with 6-18 sensors, and the dataset includes over 100k synchronous frames from different sensors. The authors annotate 3D bounding boxes for each sensor and associate the same objects across different devices and frames. They also formulate four tasks—Monocular 3D Detection, Lidar 3D Detection, Multiple Object Tracking, and Multi-sensor Multi-object Tracking—to facilitate research in roadside perception and vehicle-infrastructure cooperation. The paper provides benchmarks for these tasks and evaluates the performance of various methods using the HoloVIC dataset. The results show that the proposed dataset and benchmarks can effectively evaluate and improve the performance of perception models in autonomous driving.