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

Authors: Huixuan Zhang, Xiaojun Wan

Published: 2025-10-27 05:37:23+00:00

AI Summary

This paper introduces UniAIDet, a unified and comprehensive benchmark for AI-generated image content detection and localization, addressing limitations of existing benchmarks in coverage of generative models and image categories. UniAIDet includes both photographic and artistic images, covers a wide range of generative models (text-to-image, image-to-image, inpainting, editing, deepfake), and provides masks for fine-grained localization. Using this benchmark, the authors conduct a comprehensive evaluation of various detection methods and analyze generalization capabilities across models and image categories, revealing significant shortcomings of current approaches.

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.

Key findings

Existing AI-generated image detection and localization methods perform poorly on the comprehensive UniAIDet benchmark, indicating they are far from mature. While detection and localization tasks generally correlate, existing detection-only methods show poor generalization across different generative models, especially partial synthesis models. Furthermore, methods exhibit performance inconsistencies between photographic and artistic image categories, highlighting a significant challenge in content generalization.

Approach

The authors introduce UniAIDet, a novel, large-scale benchmark for detecting and localizing AI-generated image content. This benchmark unifies diverse image categories (photographic and artistic) and covers a wide array of generative models, including holistic and partial synthesis models. Each partial synthetic image in the benchmark is meticulously provided with a mask to evaluate localization methods, offering a more robust foundation for future research.

Datasets

MSCOCO, NYTimes800k, WikiArt, Danbooru, FFHQ

Model(s)

UNKNOWN

Author countries

China

← Previous