BGM909: A Video-Music Dataset

Paper arXiv Code Project

Introduction

Due to the lack of high-quality open-source datasets for background music generation, we collect a new video-music dataset BGM909 based on POP909 containing 909 pieces of piano version music and corresponding well-aligned videos. It has the following advantages over previous datasets.

  1. We provide high-quality MIDI files of music, and detailed annotations such as chords, styles etc.
  2. The content of the videos aligns with the music. Specifically, we provide the official MV videos for each song music to ensure semantic coherence.
  3. We also manually edit and check the video-music pairs to ensure perfect temporal alignment.
  4. Detailed annotations for videos including fine-grained language descriptions and shot transitions are provided for further study.

Demo

Files

You can download the files of BGM909 here. We provide the url and timestamp of the original videos corresponding to the music. You can also get access to our generated features for the videos and the annotations. For the music-related annotations, please refer to the original POP909 dataset.

BibTeX

@misc{li2024diffbgm,
      title={Diff-BGM: A Diffusion Model for Video Background Music Generation}, 
      author={Sizhe Li and Yiming Qin and Minghang Zheng and Xin Jin and Yang Liu},
      year={2024},
      eprint={2405.11913},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}