Skip to content

Dataset · Vincent-HKUSTGZ

AI Image Editing Detection Dataset via VQA

Vincent-HKUSTGZ/FragFake

Image-text pairs for training multimodal models to detect image editing artifacts.

What's inside

A dataset of 10K to 100K English examples for visual-question-answering. It features four data groups—Gemini-IG, GoT, MagicBrush, and UltraEdit—each split into two difficulty levels (easy and hard). Eight separate configurations are provided for accessing splits in the interface.

What you can do with it

  • Fine-tuning VLM models to detect generative edits and deepfakes in photos
  • Benchmarking AI editing detectors on simple and complex examples
Tasks
Visual question answering
Languages
English
Size
10K–100K rows
Downloads this month
20.9K
Likes
1

How to load it

from datasets import load_dataset

ds = load_dataset("Vincent-HKUSTGZ/FragFake")