Datasets

Semi-synthetic human pose dataset from top-view using NeRF

Real-world indoor dataset with 4,800 omnidirectional images, bounding boxes and keypoint annotations for persons in 15 actions.

Dataset version of THEOStereo with disparity maps for training and evaluation of stereo vision techniques based on omnidirectional images.

Synthetic indoor dataset with omnidirectional images, depth maps, bounding boxes and segmentation masks for certain object types and persons, as well as keypoints and actions of persons.

Real-world indoor dataset with omnidirectional images, bounding boxes and keypoint annotations for persons.

OmniFlow is a new synthetic omnidirectional human optical flow dataset and consists of a naturalistic 3D indoor environment with textured rooms, characters, actions, objects, illumination and motion blur. The dataset contains rendered images of household activities and corresponding forward and backward optical flow.

Synthetic indoor dataset for training and evaluation of stereo vision techniques based on omnidirectional images.

Synthetic indoor dataset with omnidirectional images, bounding boxes and segmentation masks for certain object types and persons.

Real-world indoor dataset with omnidirectional images, bounding boxes and segmentation masks for certain object types and persons.