LAION-5B
Five billion image-text pairs collected from the public web.
Resumen
LAION-5B is a collection of image and alt-text pairs, filtered by a CLIP model into subsets of increasing quality. It solves the problem that image-text training data cannot be labelled by hand at the scale contrastive training needs. It exists as the corpus open vision-language models are trained on, and its filtering by a model the data later trains is the circularity its critics point at.
Grafo de tecnologías
Cómo se conecta LAION-5B con el resto del stack.
Aplicado a
- CLIP model Multimodal LAION-5B used for CLIP High ✓ source
- Contrastive Learning technique Algorithms LAION-5B used for Contrastive Learning High ✓ source
- Multimodal AI application Multimodal LAION-5B used for Multimodal AI High ✓ source
- Pretraining technique Techniques LAION-5B used for Pretraining High ✓ source