LAION drops massive open video dataset with 10 million hours of footage for AI research

LAION has introduced the Big Video Dataset (BVD), a substantial resource for AI research comprising 80 million videos and 10 million hours of runtime. This dataset includes 55 million auto-described clips, making it one of the largest open video datasets available. The release aims to facilitate advancements in video understanding and related AI applications.

Models trained on the BVD have demonstrated improved performance, surpassing the previous benchmark set by InternVid by up to 2.1 percentage points. This enhancement in model accuracy highlights the dataset’s potential to drive significant progress in the field of video analysis and machine learning.

Legally, LAION’s ability to compile this dataset is supported by a 2024 Hamburg court ruling, which permits the collection of copyrighted content for non-commercial research purposes. This legal framework may encourage further exploration and utilization of large-scale datasets in AI research. For more details, refer to the original article on The Decoder.