1B image-text pairs with re-captions
Large DataComp-1B image-text corpus with improved captions from LLaVA-1.5-LLaMA3-8B.
- Zero shot classification
- Text retrieval
- Image → text
Hugging Face datasets with plain-language descriptions: what's inside and what models to train on them. 4,788 in the catalog so far, with new ones added every hour.
Answered by Gemini 4 · the chat is kept only in this tab
Large DataComp-1B image-text corpus with improved captions from LLaVA-1.5-LLaMA3-8B.
Million-scale dataset of text prompts and generated videos for text-to-video research.
7.8k videos as latent arrays and CLIP prompt embeddings for video generation.
Unofficial mirror of the RainDrop (CVPR 2018) dataset for training models to clean lens water drops.
Medical brain MRI k-space dataset of healthy volunteers for image reconstruction and motion artifact suppression.
Panoramic RGB-D data with camera poses for evaluating geometric coverage quality of 3D scenes.
25,435 triples of scientific papers, posters, and HTML/CSS templates from AI conferences for poster generation.
Synthetic dataset from Stability AI and LAION with English handwritten text images for OCR and generation.
Official dataset for interactive image editing based on text instructions and spatial cues.
Large corpus of before-and-after pairs with English and Chinese text instructions and quality checks.
Source image, instruction, and edited image pairs for training and evaluating image-to-image models.
A set of 3,506 examples for evaluating models that edit text and tables on document scans via instructions.
A large-scale image-text dataset with camera geometry for spatial AI.
10K triplets of source image, instruction, and result for training editing.
English expert preference dataset for training image generation reward models.
Over 4,000 carefully curated PBR materials for texture generation and computer graphics tasks.
Largest open human preference dataset for text-to-image models with 7.7M votes.
Over 5 million anime images with detailed annotations (averaging 30 tags per image) for generative models.
81,444 art images with artist, genre, and style labels.
Large-scale corpus of image-text pairs where caption phrases are grounded to corresponding image regions.
A large geospatial dataset with temporally and spatially aligned imagery from five satellite sensors.
Around one million "before/after" image pairs with text instructions for mask-free image editing.
A high-quality and diverse dataset of approximately 2 million text-image pairs for text-to-image training.
Between 1 and 10 thousand images from the famous Adobe FiveK set converted to lossless WebP to save space.