Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GenDet: Towards Good Generalizations for AI-Generated Image Detection

Mingjian Zhu, Hanting Chen, Mouxiao Huang, Wei Li, Hailin Hu, Jie Hu, Yunhe Wang

TL;DR

GenDet tackles the challenge of detecting AI-generated images from unseen generators by reframing detection as anomaly-like discrimination between real and fake inputs. It combines an adversarial teacher-student discrepancy-aware learning framework with a generalized feature augmenter, training to minimize real-input discrepancy while maximizing fake-input discrepancy, and to further generalize via adversarial feature augmentation. The approach achieves state-of-the-art results on UniversalFakeDetect and GenImage, including strong performance on unseen generators, degraded-image robustness, and cross-dataset transfer. This yields a practically impactful detector that generalizes beyond known generators and remains resilient under common image distortions.

Abstract

The misuse of AI imagery can have harmful societal effects, prompting the creation of detectors to combat issues like the spread of fake news. Existing methods can effectively detect images generated by seen generators, but it is challenging to detect those generated by unseen generators. They do not concentrate on amplifying the output discrepancy when detectors process real versus fake images. This results in a close output distribution of real and fake samples, increasing classification difficulty in detecting unseen generators. This paper addresses the unseen-generator detection problem by considering this task from the perspective of anomaly detection and proposes an adversarial teacher-student discrepancy-aware framework. Our method encourages smaller output discrepancies between the student and the teacher models for real images while aiming for larger discrepancies for fake images. We employ adversarial learning to train a feature augmenter, which promotes smaller discrepancies between teacher and student networks when the inputs are fake images. Our method has achieved state-of-the-art on public benchmarks, and the visualization results show that a large output discrepancy is maintained when faced with various types of generators.

GenDet: Towards Good Generalizations for AI-Generated Image Detection

TL;DR

GenDet tackles the challenge of detecting AI-generated images from unseen generators by reframing detection as anomaly-like discrimination between real and fake inputs. It combines an adversarial teacher-student discrepancy-aware learning framework with a generalized feature augmenter, training to minimize real-input discrepancy while maximizing fake-input discrepancy, and to further generalize via adversarial feature augmentation. The approach achieves state-of-the-art results on UniversalFakeDetect and GenImage, including strong performance on unseen generators, degraded-image robustness, and cross-dataset transfer. This yields a practically impactful detector that generalizes beyond known generators and remains resilient under common image distortions.

Abstract

The misuse of AI imagery can have harmful societal effects, prompting the creation of detectors to combat issues like the spread of fake news. Existing methods can effectively detect images generated by seen generators, but it is challenging to detect those generated by unseen generators. They do not concentrate on amplifying the output discrepancy when detectors process real versus fake images. This results in a close output distribution of real and fake samples, increasing classification difficulty in detecting unseen generators. This paper addresses the unseen-generator detection problem by considering this task from the perspective of anomaly detection and proposes an adversarial teacher-student discrepancy-aware framework. Our method encourages smaller output discrepancies between the student and the teacher models for real images while aiming for larger discrepancies for fake images. We employ adversarial learning to train a feature augmenter, which promotes smaller discrepancies between teacher and student networks when the inputs are fake images. Our method has achieved state-of-the-art on public benchmarks, and the visualization results show that a large output discrepancy is maintained when faced with various types of generators.
Paper Structure (23 sections, 5 equations, 3 figures, 7 tables)

This paper contains 23 sections, 5 equations, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. AI-Generated Image Detection
  4. Teacher-Student Anomaly Detection
  5. Method
  6. Problem Definition and Method Overview
  7. Teacher-Student Discrepancy-Aware Learning
  8. Generalized Feature Augmentation
  9. Experiments
  10. Datasets and Experimental Settings
  11. Datasets.
  12. Implementation Details.
  13. Evaluation Metrics.
  14. GenDet on UniversalFakeDetect Dataset
  15. Comparision to existing methods.
  16. ...and 8 more sections

Figures (3)

  • Figure 1: We use t-SNE to visualize real and fake images on the UniversalFakeDetect dataset ojha2023towards. The real images are shown in red. The fake images generated by seen and unseen generators are shown in orange and green, respectively. A closer distribution of real and fake images increases the difficulty of classification.
  • Figure 2: Adversarial teacher-student discrepancy-aware framework. These three stages are trained in rotation. The feature extractor and the teacher network are fixed after pretraining. The student network and feature augmenter are trained in an adversarial manner.
  • Figure 3: T-SNE degradation of GenDet (top) and CNNDet (bottom). The real images are shown in red. The fake images are shown in green, respectively. A large distribution discrepancy between real and fake images decreases the difficulty of classification.