Skip to content

Latest commit

 

History

History
79 lines (52 loc) · 4.58 KB

README.md

File metadata and controls

79 lines (52 loc) · 4.58 KB
Raw

PianoMotion10M

Dataset and Benchmark for Hand Motion Generation in Piano Performance

Qijun Gan1, Song Wang1, Shengtao Wu2 ,Jianke Zhu1 📧

1 Zhejiang University, 2 Hangzhou Dianzi University

(📧) corresponding author.

ArXiv Preprint      Project Page      Dataset[Google Drive]/[Hyper.ai]/[zenodo]

News

  • Jun. 14th, 2024: Paper is available at arxiv. ☕️
  • Jun. 1st, 2024: We released our code and datasets! Paper is coming soon. Please stay tuned! ☕️

Abstract

Recently, artificial intelligence techniques for education have been received increasing attentions, while it still remains an open problem to design the effective music instrument instructing systems. Although key presses can be directly derived from sheet music, the transitional movements among key presses require more extensive guidance in piano performance. In this work, we construct a piano-hand motion generation benchmark to guide hand movements and fingerings for piano playing. To this end, we collect an annotated dataset, PianoMotion10M, consisting of 116 hours of piano playing videos from a bird's-eye view with 10 million annotated hand poses. We also introduce a powerful baseline model that generates hand motions from piano audios through a position predictor and a position-guided gesture generator. Furthermore, a series of evaluation metrics are designed to assess the performance of the baseline model, including motion similarity, smoothness, positional accuracy of left and right hands, and overall fidelity of movement distribution. Despite that piano key presses with respect to music scores or audios are already accessible, PianoMotion10M aims to provide guidance on piano fingering for instruction purposes.

Introduction

PianoMotion10M is a large-scale piano-motion dataset. And we present a benchmark for hand motion generation with piano music.

framework

Overview of our framework. We collect videos of expert piano performances from the internet and annotated and processed them to obtain a large-scale dataset, PianoMotion10M, comprising piano music and hand motions. Building upon this dataset, we establish a benchmark aimed at generating hand movements from piano music.

Models

Sample results of our generation model.

sample_out_1.mp4
sample_out_2.mp4
sample_out_3.mp4

Results from the PianoMotion10M paper

comparison

Method Backbone Decoder FID PARAMs
EmoTalk HuBert Transformer 4.645 308
LivelySpeaker HuBert Transformer 4.157 321
Our-Base Wav2Vec2.0 SSM 3.587 320
Our-Base Wav2Vec2.0 Transformer 3.608 323
Our-Base HuBert SSM 3.412 320
Our-Base HuBert Transformer 3.529 323
Our-Large Wav2Vec2.0 SSM 3.453 539
Our-Large Wav2Vec2.0 Transformer 3.376 557
Our-Large HuBert SSM 3.395 539
Our-Large HuBert Transformer 3.281 557

Notes:

  • All the experiments are performed on 1 NVIDIA GeForce RTX 3090Ti GPU.

Getting Started

Citation

If you find PianoMotion10M is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.

@inproceedings{gan2024pianomotion,
  title={PianoMotion10M: Dataset and Benchmark for Hand Motion Generation in Piano Performance},
  author={Gan, Qijun and Wang, Song and Wu, Shengtao and Zhu, Jianke},
  year={2024},
}