LAION Launches BVD, 10 Million Hours of Free Video for Research
LAION has launched the Big Video Dataset, touted as one of the largest open video datasets dedicated to AI. The download includes 80 million videos totaling 10 million hours, with 55 million clips automatically annotated and 300 million images. Access is free and reserved for research purposes, with a possible reference to a ruling in 2024 in Hamburg.
Free Access, Research Use, and Support from a 2024 Ruling
The dataset and code are offered as open access, but strictly for research purposes. A ruling from the Hamburg Regional Court in 2024 allowed LAION to collect protected content for non-commercial use, and the organization may likely refer to this framework. LAION also asks users to respect the rights of original creators.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
A Massive Corpus from CommonCrawl and Predominantly English-Speaking
The BVD, presented as one of the largest open datasets for AI, is based on 1.3 billion video URLs found in CommonCrawl. LAION downloaded 80 million of these, amounting to 10 million hours, and extracted 55 million clips with automatically generated descriptions and 300 million still images. Most of the videos come from YouTube, and the majority are in English.
Claimed Benchmarks and Multimodal Training Methodology
According to a study, models trained on this corpus outperform comparable models trained on InternVid by up to 2.1 percentage points on common video-text evaluations. The training combines video streams, audio, and text to learn the relationships between scenes, descriptions, and sounds.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.