Background Image Background Image Background Image

Community Forensics: Using Thousands of Generators to Train Fake Image Detectors
2024

 
 

Abstract

 

One of the key challenges of detecting AI-generated images is spotting images that have been created by previously unseen generative models. We argue that the limited diversity of the training data is a major obstacle to addressing this problem, and we propose a new dataset that is significantly larger and more diverse than prior work. As part of creating this dataset, we systematically download thousands of text-to-image latent diffusion models and sample images from them. We also collect images from dozens of popular open source and commercial models. The resulting dataset contains 2.7M images that have been sampled from 4803 different models. These images collectively capture a wide range of scene content, generator architectures, and image processing settings. Using this dataset, we study the generalization abilities of fake image detectors. Our experiments suggest that detection performance improves as the number of models in the training set increases, even when these models have similar architectures. We also find that increasing the diversity of the models improves detection performance, and that our trained detectors generalize better than those trained on other datasets.


The Community Forensics Dataset

 
Your browser does not support SVG
 

We sample images from 4803 generators, approximately 250 times more than previous datasets. Our dataset contains 2.7M images generated by: (a) thousands of systematically downloaded open-source text-to-image latent diffusion models, (b) manually chosen open-source models with a variety of architectures, and (c) state-of-the-art commercial models. We use this dataset to study generalization in the generated image detection problem.


Motivation

 
Your browser does not support SVG
 

Training from more generators improves generalization across architectures. Although the classifiers in the above figure only train on latent diffusion models, we observe an increase in performance on other architectures such as GANs and pixel-based diffusion models as we add more models.


Results

 
Your browser does not support SVG
 

Classifiers trained on our dataset obtain strong performance, both on our newly proposed evaluations and on multiple existing benchmarks.


Your browser does not support SVG
 

Our classifiers performs well across various architectures. We observe that our classifiers generalize better across architectures compared to those trained on other datasets.


Citation

 

The website is based on a template by Ref-NeRF and Michaël Gharbi.