LAION unleashes 10m hour video dataset for AI research
laion releases big video dataset with 10 million hours of footage, outperforming internvid on benchmarks. this impacts ai research, particularly in video-to-text models.
laion has released the big video dataset (bvd), a massive open video dataset for ai research, containing 10 million hours of footage from 80 million videos. the dataset was created from 1.3 billion video urls found in commoncrawl, with most videos coming from youtube and the majority in english. the team extracted 55 million clips with auto-generated video and audio descriptions, plus 300 million still images. models trained on bvd have been shown to outperform comparable models trained on internvid by up to 2.1 percentage points on common video-to-text benchmarks. this development impacts ai research, particularly in video-to-text models, and may influence future advancements in the field.