Synthetic multi-camera temporal dataset collected in CARLA Town04 for
StreamPETR 3D object detection experiments.
8,100 synchronized frames and 48,600 JPEG images
Six cameras per frame at 1600x900
405 disjoint temporal clips
Nine weather presets
Three maneuvers: left, right, and straight
27 weather-maneuver cases with 300 frames per case
139,425 labeled 3D objects
Classes present: car, truck, bus, motorcycle, and bicycle… See the full description on the dataset page:
https://huggingface.co/datasets/BanVienCorp/StreamPETR_3D_Dataset.