README.md

metadata

license: cc-by-4.0
task_categories:
  - other
tags:
  - audio-visual-sound-source-localization
  - multimodal-data
  - synthetic-data
  - multi-modal
  - cvpr-2026
  - stable-diffusion-3
  - stable-audio
pretty_name: VGGSynth1

VGGSynth1: Synthetic Audio-Visual Dataset (Part 2)

How Far Can We Go With Synthetic Data for Audio-Visual Sound Source Localization? (CVPR 2026 Highlight)

Authors

Arda Senocak*, Sooyoung Park*, Tae-Hyun Oh, Joon Son Chung (* Equal Contribution)

---

Introduction

VGGSynth2 is a high-fidelity synthetic clone of the VGGSound dataset, built using state-of-the-art generative models. This dataset is designed to explore the boundaries and utility of synthetic data in training models for Audio-Visual Sound Source Localization (SSL).

• Visuals: Generated via Stable Diffusion 3 (SD3)
• Audio: Generated via Stable Audio

Citation

@inproceedings{senocak2026howfar,
  title={How Far Can We Go With Synthetic Data for Audio-Visual Sound Source Localization?},
  author={Senocak, Arda and Park, Sooyoung and Oh, Tae-Hyun and Chung, Joon Son},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2026}
}