OmniDirectional INdoor (ODIN) dataset

The OmniDirectional INdoor (ODIN) dataset is a large-scale omnidirectional dataset that contains a wide range of synchronized modalities, including images and videos captured by cameras of different types while participants engage in various daily activities, as well as their physiological data. The ODIN dataset will enable research in diverse areas such as human pose estimation, activity recognition, person tracking and monitoring, scene understanding, privacy preservation, biometric monitoring, novel view synthesis, generative modeling, 3D scene reconstruction, and image registration. With our initial release, we aim to stimulate research in 3D human pose estimation using omnidirectional cameras. This area of research has been limited, perhaps due to the difficulty of the problem and the scarcity of datasets. ODIN provides camera-frame 3D pose estimates associated with omnidirectional camera images. We propose an innovative unsupervised pipeline to obtain these pose estimates in real-life indoor environments, while maintaining the state of the surroundings, and without the need for expensive equipment. The dataset is accessible at https://web.ua.es/en/ami4aha/odin-dataset.html.