Beyond Binary Classification: A Semi-supervised Approach to Generalized AI-generated Image Detection

2025-11-23

AI Summary

Existing AI-generated image detectors exhibit poor generalization across diverse generative architectures (e.g., GANs vs. diffusion models). Method: This paper presents the first theoretical analysis of architectural differences from a manifold coverage perspective and proposes TriDetect—a novel tripartite detection framework that achieves fine-grained discrimination and cross-architecture generalization by uncovering latent generation-architecture patterns embedded in forged images. It introduces a Sinkhorn-Knopp-based balanced clustering assignment and a cross-view consistency mechanism, enabling semi-supervised discovery and modeling of implicit generative model categories without architecture-level labels. Contribution/Results: Evaluated on two standard benchmarks and three real-world datasets, TriDetect consistently outperforms 13 state-of-the-art baselines, achieving substantial gains in zero-shot detection generalization to unseen generators.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Neural Architectures and Foundation Models
• Natural Language Processing: Generation

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

The rapid advancement of generators (e.g., StyleGAN, Midjourney, DALL-E) has produced highly realistic synthetic images, posing significant challenges to digital media authenticity. These generators are typically based on a few core architectural families, primarily Generative Adversarial Networks (GANs) and Diffusion Models (DMs). A critical vulnerability in current forensics is the failure of detectors to achieve cross-generator generalization, especially when crossing architectural boundaries (e.g., from GANs to DMs). We hypothesize that this gap stems from fundamental differences in the artifacts produced by these extbf{distinct architectures}. In this work, we provide a theoretical analysis explaining how the distinct optimization objectives of the GAN and DM architectures lead to different manifold coverage behaviors. We demonstrate that GANs permit partial coverage, often leading to boundary artifacts, while DMs enforce complete coverage, resulting in over-smoothing patterns. Motivated by this analysis, we propose the extbf{Tri}archy extbf{Detect}or (TriDetect), a semi-supervised approach that enhances binary classification by discovering latent architectural patterns within the "fake" class. TriDetect employs balanced cluster assignment via the Sinkhorn-Knopp algorithm and a cross-view consistency mechanism, encouraging the model to learn fundamental architectural distincts. We evaluate our approach on two standard benchmarks and three in-the-wild datasets against 13 baselines to demonstrate its generalization capability to unseen generators.

Problem

• Detecting AI-generated images across different architectural families
• Addressing cross-generator generalization failure in image forensics
• Developing semi-supervised detection that identifies latent architectural patterns

Innovation

• Semi-supervised approach for AI-generated image detection
• TriDetect discovers latent architectural patterns in fake images
• Uses balanced clustering and cross-view consistency mechanisms