LAION releases 10M-hour open video dataset LAION-BVD for multimodal pre-training
thione · x · 2026-08-31
LAION released LAION-BVD, a large-scale open multimodal training dataset containing 80M videos totaling 10 million hours. Sourced from 1.3B URLs via Common Crawl, it is designed for video, audio, and image pre-training. Using content-aware scene detection, researchers extracted clips and generated synthetic captions. Models trained on this data achieve competitive performance on standard video-text and audio-text benchmarks, with consistent scaling improvements.
More from Research
- RL Experts Surprised by Agents Voluntarily Self-Destructing to Aid Peers — ZeroStateReflex · 2026-08-31
- Milestone: pLM-designed peptides work in vivo, selectively degrading β-catenin in mice — arjunrajlab · 2026-08-31
- Fast Sim2Real Workflow: Mobile Policy Viewer and Parameter Sweeping — yacineMTB · 2026-08-31
- Wilcoxon signed rank test outperforms McNemar for binary data evals — IanArawjo · 2026-08-31
- Chollet responds to ARC-AGI eval dispute: don't claim untested scores — fchollet · 2026-08-31
- Research复盘：Linear attention found ineffective in specific setup — jm_alexia · 2026-08-31