This dataset contains raw video and music files from the MGSV-EC benchmark. Due to copyright and redistribution restrictions (content sourced from Kuaishou and third-party creators), access is granted only for non-commercial academic research purposes. By requesting access, you agree to the following terms.

MGSV-EC Raw Media

Raw video and music files for the MGSV-EC benchmark, accompanying the paper:

Music Grounding by Short Video (ICCV 2025)

The official repository at xxayt/MGSV provides annotations, pre-extracted features, and code. This repository supplements it with the original media files to enable research on new video-audio grounding, encoder, or end-to-end methods.

Access

This is a gated dataset with manual review. Request access above.

Data Coverage

The full MGSV-EC benchmark contains 53,194 videos paired with 4,050 unique music tracks. This repository provides a partial release of the raw media due to source platform restrictions:

Type	Files in this repo	Full dataset
Music tracks	7,580 (incl. vocal/no-vocal variants)	4,050 unique tracks
Test videos	1,721	2,000
Train/Val videos	2,085	51,194

Pre-extracted features (ViT-B-32 CLIP visual + AST audio) for all 53,194 entries are available at xxayt/MGSV-EC. For reproducing the benchmark results reported in the paper, the pre-extracted features are sufficient.

Files

Directory	File	Size	Description
`raw_music/`	`music.tar.gz.00`	25 GB	All music tracks (split archive, 1 part)
`raw_videos/`	`test_videos.tar.gz.{00,01,02}`	109 GB	Test set videos (split archive, 3 parts)
`raw_videos/`	`trainval_videos.tar.gz.{00,01,02}`	128 GB	Train/val set videos (split archive, 3 parts)
Total	7 files	~261 GB

Download & Extract

# Download via huggingface-cli
huggingface-cli download xxayt/MGSV-EC-Raw --repo-type dataset --local-dir MGSV-EC-Raw

# Extract music
cd MGSV-EC-Raw/raw_music
cat music.tar.gz.* | tar -xzf -

# Extract test videos
cd ../raw_videos
cat test_videos.tar.gz.* | tar -xzf -

# Extract train/val videos
cat trainval_videos.tar.gz.* | tar -xzf -

Terms of Use

Non-commercial academic research only. Commercial use requires contacting the authors.
No redistribution. Do not share the downloaded files or any derived copies.
Citation required. Please cite the MGSV paper in any published work.
Licensed under CC BY-NC-ND 4.0.

Citation

@inproceedings{xin2025mgsv,
  title={Music Grounding by Short Video},
  author={Xin, Zijie and Wang, Minquan and Liu, Jingyu and Chen, Quan and Ma, Ye and Jiang, Peng and Li, Xirong},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  year={2025}
}

Related Resources

Code & annotations: github.com/xxayt/MGSV
Pre-extracted features: huggingface.co/datasets/xxayt/MGSV-EC

Downloads last month: 60

Total file size:

280 GB

Collection including xxayt/MGSV-EC-Raw

[ICCV2025]MGSV

Collection

[ICCV 2025] Music Grounding by Short Video • 5 items • Updated 5 days ago • 1