For the CVPR 2023 WoodScape Challenge, Parallel Domain created a dataset of 106 synthetic urban scenes across 4,240 keyframes, captured under variable cloudiness. Each keyframe, recorded via four fisheye cameras, includes 2D and 3D bounding boxes, depth, semantic, and motion segmentation annotations, totaling 16,960 annotated images paired with prior RGB frames.
The Open Cyclist Synthetic Dataset offers a collection of over 78,750 RGB images, with 2D bounding box annotations. It is designed to improve the detection of rare classes such as cyclists in autonomous systems, tackling the ‘long tail problem’ faced by such systems.