Cityscapes 3D Dataset Released

Cityscapes 3D is an extension of the original Cityscapes with 3D bounding box annotations for all types of vehicles as well as a benchmark for the 3D detection task. For more details please refer to our paper, presented at the CVPR 2020 Workshop on Scalability in Autonomous Driving.

Today, we released our 3D bounding box annotations of all vehicle types, i.e. car, truck, bus, on rails, motorcycle, bicycle, caravan, and trailer. The box annotations feature a full 3D orientation including yaw, pitch, and roll labels. The annotations are available on our download page.

Our toolbox supports the new annotations and is available on Github or can be installed using pip, i.e.python -m pip install cityscapesscripts[gui]. In order to visualize the 3D Boxes, run csViewer and select the CS3D ground truth. The toolbox also includes our evaluation code, run csEvalObjectDetection3d -h for details. You can evaluate your method on the validation set as of today. We plan to launch our benchmark as well as the evaluation server in time for the CVPR2021 submission deadline.

The tool csViewer can be used to visualize the 3D bounding box annotations.