RehabU-15-Skeleton: A De-identified Kinematic Dataset for Upper Limb Rehabilitation Assessment

This dataset contains the completely de-identified 3D skeleton coordinate features used for upper limb rehabilitation assessment. To comply with hospital privacy regulations, no raw RGB videos are included.The dataset includes 15 classes of upper limb rehabilitation actions. The kinematic sequences are processed into a decoupled fine-grained spatiotemporal graph (DF-SG) topology, encompassing 51 nodes that capture both major trunk movements and fine-grained hand compensatory features.Data structure includes normalized coordinate tensors and matching clinical ground-truth scores evaluated by medical experts.