AudioVisual-Caption/

ASID-1M

[🏠 Homepage] [📖 Arxiv Paper] [🤗 Models & Datasets] [💻 Code]

We introduce ASID-1M, a large-scale audiovisual instruction dataset built to support universal video understanding with fine-grained, controllable supervision. Most existing video-instruction data represents complex audiovisual content as a single, monolithic caption. This often leads to incomplete coverage (missing audio… See the full description on the dataset page: https://huggingface.co/datasets/AudioVisual-Caption/ASID-1M.