You need to agree to share your Hugging Face information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

FlatSounds is released under CC BY-NC 4.0: you may use it for non-commercial purposes only, with attribution.

Log in or Sign Up to review the conditions and access this dataset content.

FlatSounds

Videos and annotations of FlatSounds, a benchmark for physical video-to-audio generation (CVPR 2026). The evaluation code, documentation and leaderboard are at github.com/Tinglok/flatsounds.

Path Content
benchmark_warped_clips/ 542 FlatSounds-Physics videos: the paper split uses 446, the verified split 462. In a video pair, the counterfactual video is time-warped so that its hits occur at the same times as in the factual video
benchmark_clips_full/ 185 FlatSounds-Single videos: the original recordings
benchmark/physics_verified.json FlatSounds-Physics, verified split (default): 302 test cases, 724 tests
benchmark/physics_paper.json FlatSounds-Physics, paper split: 268 test cases, 620 tests
benchmark/single.json FlatSounds-Single clips: video, caption, duration and hit times

The videos are 2.5-10 s long and keep their recorded sound: give your model only the video frames.

Download

Accept the access conditions on this page, log in with hf auth login, then:

hf download Tinglok/FlatSounds --repo-type dataset --local-dir FlatSounds

The GitHub README explains how to generate outputs and evaluate them.

License

The data is released under CC BY-NC 4.0; the evaluation code is MIT.

Citation

@inproceedings{li2026benchmarking,
  title={Benchmarking Single-Factor Physical Video-to-Audio Generation},
  author={Li, Tingle and Gururani, Siddharth and Shih, Kevin J. and Bhatt, Gantavya and Lee, Sang-gil and Kong, Zhifeng and Goel, Arushi and Anumanchipalli, Gopala and Liu, Ming-Yu},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2026}
}
Downloads last month
11