Dataset · ComplexDataLab
Deepfake and AI photo detection dataset and benchmark
ComplexDataLab/OpenFake
Millions of real and generated image pairs on socio-political topics for training and testing AI content detectors.
What's inside
Contains between 1M and 10M images with English text data, pairing real photographs with their synthetic copies. Fakes are generated using proprietary models, open diffusion neural networks, and custom fine-tunes, with a focus on politically and socially significant subjects. To test detector performance in real-world conditions, the set also includes a separate test slice with images from Reddit.
What you can do with it
- Training classifiers to identify neural-network-generated images
- Testing moderation system robustness against modern generators and deepfakes
- Tasks
- Image classification
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 29.1K
- Likes
- 43
Models trained on it
- bendavidsteel/openfake-swinv2-baseImage classification 325
- bendavidsteel/openfake-swinv2-base-textgraphics-loraImage classification 130
- bendavidsteel/openfake-swinv2-base-attribution-headImage classification 129
- husseinelsaadi/aidetect-vit-b16Image classification 0
- GPTchatly/slope-scopeImage classification 0
- s-gehring/detectra-openfakeImage classification 0
How to load it
from datasets import load_dataset
ds = load_dataset("ComplexDataLab/OpenFake")