Towards Universal Fake Image Detectors that Generalize Across Generative Models

By Utkarsh Ojha, Yuheng Li, Yong Jae Lee. Computer Vision and Pattern Recognition, 2023

644 citations recorded by the source corpus.

With generative models proliferating at a rapid rate, there is a growing need for general purpose fake image detectors. In this work, we first show that the existing paradigm, which consists of training a deep network for real-vs-fake classification, fails to detect fake images from newer breeds of generative models when trained to detect GAN fake images. Upon analysis, we find that the resulting classifier is asymmetrically tuned to detect patterns that make an image fake. The real class becomes a ‘sink’ class holding anything that is not fake, including generated images from models not accessible during training. Building upon this discovery, we propose to perform real-vs-fake classification without learning; i.e., using a feature space not explicitly trained to distinguish real from fake images. We use nearest neighbor and linear probing as instantiations of this idea. When given access to the feature space of a large pretrained vision-language model, the very simple baseline of nearest neighbor classification has surprisingly good generalization ability in detecting fake images from a wide variety of generative models; e.g., it improves upon the SoTA [50] by +15.07 mAP and +25.90% acc when tested on unseen diffusion and autoregressive models. Our code, models, and data can be found at https://github.com/Yuheng-Li/UniversalFakeDetect

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/0fafac05423aa187d85cc12bce1865aefc787700

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Frequency-Aware Deepfake Detection: Improving Generalizability through Frequency Space Domain Learning (2024, background): “It is evident that our Fre-qNet, with a modest parameter count of 1.9 million, significantly outperforms the current state-of-the-art model Ojha (Ojha et al. 2023), which boasts an extensive 304 million parameters.”
Understanding Generative AI Content with Embedding Models (2024, methodology): “Others have used embedding models trained via CLiP to distinguish real from synthetic images by applying various classifiers to the resulting embeddings [24–26].”