drhead/ZeroDiffusion
Text-to-Image • Updated • 17
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
A subset of LAION Aesthetics v2 5+, filtered to include only high resolution (1024x1024+) images, then processed to remove dead links (as of October 2023), and with embeddings recalculated using CLIP-ViT-L-patch14 which were used to both remove poor-quality image-caption matches (CLIP similarity <0.2) and to deduplicate the dataset. Roughly one third of the dataset was dropped by these operations after filtering high resolution images.