Supplementary Materials: Humans in Kitchens: A Dataset for Multi-Person Human Motion Forecasting with Scene Context

Figure 1: Sample scenes with 3d human poses projected onto camera views for each kitchen. The first 24 joints are equivalent to the SMPL pose [1] while joints 25 to 28 represent the head, with 24 being the nose, 25 and 26 being the eyes and 27 and 28 being the ears, similar to the OpenPose skeleton. Our dataset contains two folders, poses and scenes. Each file represents the entire motion from when a person enters the scene until they leave. As persons may enter and leave multiple times a new file is created whenever they re-enter the scene. The person can be uniquely identified by their pid while the sequence number tracks the amount of times they have re-entered the scene.