This project utilizes the TUMTraf V2X Cooperative Perception Dataset, created for cooperative 3D object detection and tracking tasks.
We have selected 9 scenarios, each containing 50 frames, for tasks such as scene reconstruction and novel view synthesis. The selected scenarios are as follows:
| Scene Name |
|---|
| Ego Vehicle Occlusion |
| Cyclist in Blind Spot |
| Pedestrians Crossing |
| U-Turn Maneuver |
| RSU Occlusion |
| Far Distance Pedestrian |
| Dense VRU Crossing |
| Dense VRU with RSU Occlusion |
| Night Scene |
The dataset includes data from 2 road-side unit (RSU) cameras, 1 camera in the ego vehicle, and LiDARs whose fields of view (FOV) collectively cover the intersection. Below is the mapping of sensor and LiDAR names to their simplified identifiers:
| Sensor Name | Identifier |
|---|---|
s110_camera_basler_south2_8mm |
cam01 |
vehicle_camera_basler_16mm |
cam02 |
s110_camera_basler_south1_8mm |
cam03 |
s110_lidar_ouster_south_and_vehicle_lidar_robosense_registered |
lidar |
To preprocess the dataset as described above, run the following command:
python3 data_preprocessor.py --dataset "path_to_dataset"