Your AI-Generated Image Detector Can Secretly Achieve SOTA Accuracy, If Calibrated

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Yang, Muli, Goenawan, Gabriel James, Wang, Henan, Qin, Huaiyuan, Xu, Chenghao, Yang, Yanhua, Fang, Fen, Sun, Ying, Lim, Joo-Hwee, Zhu, Hongyuan
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908805389877248
author Yang, Muli
Goenawan, Gabriel James
Wang, Henan
Qin, Huaiyuan
Xu, Chenghao
Yang, Yanhua
Fang, Fen
Sun, Ying
Lim, Joo-Hwee
Zhu, Hongyuan
author_facet Yang, Muli
Goenawan, Gabriel James
Wang, Henan
Qin, Huaiyuan
Xu, Chenghao
Yang, Yanhua
Fang, Fen
Sun, Ying
Lim, Joo-Hwee
Zhu, Hongyuan
contents Despite being trained on balanced datasets, existing AI-generated image detectors often exhibit systematic bias at test time, frequently misclassifying fake images as real. We hypothesize that this behavior stems from distributional shift in fake samples and implicit priors learned during training. Specifically, models tend to overfit to superficial artifacts that do not generalize well across different generation methods, leading to a misaligned decision threshold when faced with test-time distribution shift. To address this, we propose a theoretically grounded post-hoc calibration framework based on Bayesian decision theory. In particular, we introduce a learnable scalar correction to the model's logits, optimized on a small validation set from the target distribution while keeping the backbone frozen. This parametric adjustment compensates for distributional shift in model output, realigning the decision boundary even without requiring ground-truth labels. Experiments on challenging benchmarks show that our approach significantly improves robustness without retraining, offering a lightweight and principled solution for reliable and adaptive AI-generated image detection in the open world. Code is available at https://github.com/muliyangm/AIGI-Det-Calib.
format Preprint
id arxiv_https___arxiv_org_abs_2602_01973
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Your AI-Generated Image Detector Can Secretly Achieve SOTA Accuracy, If Calibrated
Yang, Muli
Goenawan, Gabriel James
Wang, Henan
Qin, Huaiyuan
Xu, Chenghao
Yang, Yanhua
Fang, Fen
Sun, Ying
Lim, Joo-Hwee
Zhu, Hongyuan
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Despite being trained on balanced datasets, existing AI-generated image detectors often exhibit systematic bias at test time, frequently misclassifying fake images as real. We hypothesize that this behavior stems from distributional shift in fake samples and implicit priors learned during training. Specifically, models tend to overfit to superficial artifacts that do not generalize well across different generation methods, leading to a misaligned decision threshold when faced with test-time distribution shift. To address this, we propose a theoretically grounded post-hoc calibration framework based on Bayesian decision theory. In particular, we introduce a learnable scalar correction to the model's logits, optimized on a small validation set from the target distribution while keeping the backbone frozen. This parametric adjustment compensates for distributional shift in model output, realigning the decision boundary even without requiring ground-truth labels. Experiments on challenging benchmarks show that our approach significantly improves robustness without retraining, offering a lightweight and principled solution for reliable and adaptive AI-generated image detection in the open world. Code is available at https://github.com/muliyangm/AIGI-Det-Calib.
title Your AI-Generated Image Detector Can Secretly Achieve SOTA Accuracy, If Calibrated
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2602.01973