🤖 AI Summary
Existing AI-generated image detection methods—particularly those based on GANs and diffusion models—exhibit limited generalization to unseen generative models. Method: This paper proposes a language-guided contrastive learning framework that introduces, for the first time, a joint language–vision contrastive supervision mechanism. By leveraging textual labels to enhance visual feature learning, the approach enables zero-shot detection of images from unknown generative models. Specifically, it freezes the CLIP visual encoder and incorporates a learnable text projection head to align multimodal features, thereby improving cross-model generalization. Contribution/Results: Evaluated on four benchmark datasets, the method consistently outperforms all state-of-the-art approaches, achieving an average 12.6% improvement in detection accuracy for unseen generative models. The source code is publicly available.
📝 Abstract
The heightened realism of AI-generated images can be attributed to the rapid development of synthetic models, including generative adversarial networks (GANs) and diffusion models (DMs). The malevolent use of synthetic images, such as the dissemination of fake news or the creation of fake profiles, however, raises significant concerns regarding the authenticity of images. Though many forensic algorithms have been developed for detecting synthetic images, their performance, especially the generalization capability, is still far from being adequate to cope with the increasing number of synthetic models. In this work, we propose a simple yet very effective synthetic image detection method via a language-guided contrastive learning. Specifically, we augment the training images with carefully-designed textual labels, enabling us to use a joint visual-language contrastive supervision for learning a forensic feature space with better generalization. It is shown that our proposed LanguAge-guided SynThEsis Detection (LASTED) model achieves much improved generalizability to unseen image generation models and delivers promising performance that far exceeds state-of-the-art competitors over four datasets. The code is available at https://github.com/HighwayWu/LASTED.
