Dataset · google
DOCCI images with descriptions
google/docci
Images with detailed human descriptions and pairs of similar frames for evaluating vision models.
What's inside
A collection of images with detailed descriptions of key elements, background, lighting, and scenes. Descriptions are manually annotated and designed to distinguish each image from similar ones. The set includes related images with key differences; the description language is English.
What you can do with it
- Evaluating text-to-image models for detail accuracy
- Training image-to-text models on detailed descriptions
- Benchmarking differences between similar images
- Tasks
- Text → image, Image → text
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 889
- Likes
- 82
Models trained on it
- gokaygokay/Florence-2-SD3-CaptionerImage text → text 4.5K
- gokaygokay/sd3-long-captioner-v2Image text → text 86
- gokaygokay/sd3-long-captionerImage text → text 62
- fal/moondream2-docci-instructImage text → text 36
- gokaygokay/paligemma-rich-captionsImage text → text 25
- gokaygokay/paligemma-docci-transformersImage text → text 19
- Artiron/Florence-2-SD3-CaptionerImage text → text 13
- SoulCub/Florence-2-SD3-CaptionerImage text → text 11
How to load it
from datasets import load_dataset
ds = load_dataset("google/docci")