Abstract: Current fake image detectors trained on large synthetic image datasets perform satisfactorily on limited studied generative models. However, these detectors suffer a notable performance decline over unseen models. Besides, collecting adequate training data from online generative models is often expensive or infeasible. To overcome these issues, we propose Few-Shot Detector (FSD), a novel AI-generated image detector which learns a specialized metric space for effectively distinguishing unseen fake images using very few samples. Experiments show that FSD achieves state-of-the-art performance by $+11.6\%$ average accuracy on the GenImage dataset with only $10$ additional samples. More importantly, our method is better capable of capturing the intra-category commonality in unseen images without further training. Our code is available at https://github.com/teheperinko541/Few-Shot-AIGI-Detector.

Lay Summary:

Current tools for spotting AI-generated fakes struggle when new image generators emerge. They're only reliable for specific models they were trained on, and collecting enough training images for every new AI system is often impossible or prohibitively expensive.We developed FSD (Few-Shot Detector), a novel approach that can identify fake images from unseen AI models using just 10 sample images, with no retraining required. FSD quickly learns to recognize the hidden patterns specific to each generator by comparing image features in a specialized metric space, which allows it to instantly adapt to new generators with minimal data.As AI image generators rapidly evolve, FSD provides a practical, low-cost solution that can keep pace with emerging threats. This makes FSD an important technology for reducing cheating, misinformation, and harm on Internet platforms and social media.