swimmiing/

VGGSynth2

	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)

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… See the full description on the dataset page: https://huggingface.co/datasets/swimmiing/VGGSynth2.