The PianoMotion10M dataset is a benchmark dataset for piano gesture generation jointly constructed by Zhejiang University and Hangzhou Dianzi University in 2024. The dataset is characterized by its scale and level of detail, covering 116 hours of piano performance videos shot from a bird's-eye view and containing 10 million annotated hand gestures. These data collect a variety of musical styles and piano playing techniques, providing researchers with a rich resource to explore the relationship between piano music and hand movements.

The relevant paper results arePianoMotion10M: Dataset and Benchmark for Hand Motion Generation in Piano Performance".