UniAIDet: A Unified and Universal Benchmark for AI-Generated Image Content Detection and Localization

2025-10-27

AI Summary

Existing AI-generated image detection benchmarks suffer from narrow model coverage, limited image diversity (notably lacking artistic imagery), and inadequate support for end-to-end editing scenarios. To address these limitations, we introduce the first unified, general-purpose benchmark for both detection and localization of AI-generated images. Our benchmark systematically encompasses major generative paradigms—including text-to-image synthesis, image editing, inpainting, and deepfakes—and uniquely supports both photographic and artistic image domains. We propose a multi-scale evaluation framework that jointly measures detection accuracy and localization fidelity. Through comprehensive comparative experiments, we rigorously assess state-of-the-art methods across cross-model, cross-category, and complex end-to-end editing settings. Empirical results reveal significant performance degradation on artistic images and end-to-end editing tasks, highlighting critical gaps in current approaches. This work establishes a standardized, reproducible benchmark, provides an open analysis framework, and identifies key challenges to guide future research.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Philosophy and Ethics of AI: Artificial General Intelligence

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

With the rapid proliferation of image generative models, the authenticity of digital images has become a significant concern. While existing studies have proposed various methods for detecting AI-generated content, current benchmarks are limited in their coverage of diverse generative models and image categories, often overlooking end-to-end image editing and artistic images. To address these limitations, we introduce UniAIDet, a unified and comprehensive benchmark that includes both photographic and artistic images. UniAIDet covers a wide range of generative models, including text-to-image, image-to-image, image inpainting, image editing, and deepfake models. Using UniAIDet, we conduct a comprehensive evaluation of various detection methods and answer three key research questions regarding generalization capability and the relation between detection and localization. Our benchmark and analysis provide a robust foundation for future research.

Problem

• Evaluating AI-generated image detection across diverse models and categories
• Assessing generalization capability of detection and localization methods
• Addressing limitations in current benchmarks for comprehensive AI content analysis

Innovation

• Unified benchmark for AI-generated image detection
• Covers diverse generative models and image categories
• Evaluates generalization and localization capabilities