We introduce Shot7M2, the Synthetic, Hierarchical, and cOmpositional baskeTball dataset, consisting of 7.2 million frames designed to showcase hierarchical organization within basketball behaviors.Based on the animation models by Starke et al., we use a Neural State Machine composed of a Motion Prediction Network and a Gating Network to predict future character poses from current pose, set of control commands, goal position, environment geometry and a set of… See the full description on the dataset page:
https://huggingface.co/datasets/amathislab/SHOT7M2.
