This dataset contains about 110k images annotated with the depth and occlusion relationships between arbitrary objects. It enables research on the 2.5D Visual Relationship Detection (2.5VRD) introduced in https://arxiv.org/abs/2104.12727. - View it on GitHub