OmniDFA: A Unified Framework for Open Set Synthesis Image Detection and Few-Shot Attribution

Authors: Shiyu Wu, Shuyan Li, Jing Li, Jing Liu, Yequan Wang

Published: 2025-09-30 02:36:40+00:00

Comment: 19 pages, 5 figures

AI Summary

This paper introduces OmniDFA, a unified framework for AI-generated image (AIGI) detection and open-set, few-shot source attribution. It addresses the challenges of diverse generative models by reliably identifying unseen generators using only limited samples. To facilitate this work, the authors constructed OmniFake, a large class-aware synthetic image dataset containing 1.17 million images from 45 distinct generative models.

Abstract

AI-generated image (AIGI) detection and source model attribution remain central challenges in combating deepfake abuses, primarily due to the structural diversity of generative models. Current detection methods are prone to overfitting specific forgery traits, whereas source attribution offers a robust alternative through fine-grained feature discrimination. However, synthetic image attribution remains constrained by the scarcity of large-scale, well-categorized synthetic datasets, limiting its practicality and compatibility with detection systems. In this work, we propose a new paradigm for image attribution called open-set, few-shot source identification. This paradigm is designed to reliably identify unseen generators using only limited samples, making it highly suitable for real-world application. To this end, we introduce OmniDFA (Omni Detector and Few-shot Attributor), a novel framework for AIGI that not only assesses the authenticity of images, but also determines the synthesis origins in a few-shot manner. To facilitate this work, we construct OmniFake, a large class-aware synthetic image dataset that curates $1.17$ M images from $45$ distinct generative models, substantially enriching the foundational resources for research on both AIGI detection and attribution. Experiments demonstrate that OmniDFA exhibits excellent capability in open-set attribution and achieves state-of-the-art generalization performance on AIGI detection. Our dataset and code will be made available.

Key findings

OmniDFA achieves state-of-the-art generalization performance in AI-generated image detection, showing an average improvement of +5.83% accuracy on OmniFake and remarkable high performance on GenImage. It demonstrates strong capabilities in open-set few-shot source attribution and exhibits robust real-world applicability on the challenging Chameleon benchmark, outperforming the second-best method by +17.71% accuracy.

Approach

OmniDFA employs a dual-path architecture to capture both low-level and high-level image features for comprehensive representation. It leverages supervised contrastive learning to discriminate between various fake image sources and incorporates a sphere center loss with a learnable boundary to compactly cluster real images, enhancing authenticity detection.

Datasets

OmniFake, GenImage, Chameleon, LAION-5B, Wukong, ImageNet-1k, CC12M, MSCOCO, FFHQ, CelebA-HQ, LSUN-church, IMD2020, FODB, WildFake, MPBench, HuggingFace datasets (for specific models like DALLE3, Ideogram, Midjourney V6, Hunyuan-DiT, SD3-Medium, Janus-pro, BAGEL, Show-o, OmniGen2, Ovis-U1, UniWorld-V1)

Model(s)

UNKNOWN

Author countries

China, UK

← Previous