arXiv cs.AI· Ruoxin Chen, Jiahui Gao, Kaiqing Lin, Keyue Zhang, Yandan Zhao, Isabel Guan, Taiping Yao, Shouhong Ding·· 13 小时前AI 评分31
正交监督解耦：AlignGemini 如何实现可泛化的 AI 生成图像检测
Align, Don't Blend: Orthogonal Supervision Disentanglement As the Key to Generalizable AI-Generated Image Detection
AI 导读
研究者提出正交监督解耦（Orthogonal Supervision Disentanglement），并落地为双分支检测器 AlignGemini：VLM 只接收纯语义监督，视觉专家只接收纯伪影监督。仅用两个生成器的图像训练、不依赖复杂训练策略，该方案在五个真实场景基准上平均提升 9.5%。团队同时发布覆盖 8 个当代生成器（含 5 个闭源商用模型）的诊断基准 AIGI-Now，代码已公开。
正文 · 原文
Abstract:Recent advances in AI-Generated Image (AIGI) detection have largely focused on model architectures, loss design, and training strategies, while overlooking a more fundamental factor: how supervision is structured. Existing detectors typically train on benchmark-defined data that implicitly blends semantic and artifact signals, leaving supervision design unexplored. In this paper, we examine supervision structure through controlled experiments and show that such blended supervision is not merely suboptimal but fundamentally counterproductive. Replacing it with disentangled, orthogonal supervision - paired with inductively matched models - suffices to achieve strong cross-generator generalization. We formalize this as Orthogonal Supervision Disentanglement and instantiate it in AlignGemini, a two-branch detector in which a VLM receives purely semantic supervision and a vision expert receives purely artifact supervision. With disentangled supervision coupled with inductively matched models, even the simplest training recipe achieves a significant improvement of 9.5% across five in-the-wild benchmarks - trained on images from only two generators, without complex training strategies or diverse generator coverage. We further introduce AIGI-Now, a diagnostic benchmark covering eight contemporary generators, including five closed-source commercial models, with disentangled semantic and pixel evaluation subsets for fine-grained analysis of detector capabilities. Our code is publicly available at this https URL.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2512.06746 [cs.CV] |
| (or arXiv:2512.06746v3 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2512.06746 arXiv-issued DOI via DataCite |
Submission history
From: Ruoxin Chen [view email]
[v1]
Sun, 7 Dec 2025 09:19:00 UTC (18,834 KB)
[v2]
Fri, 30 Jan 2026 08:01:18 UTC (14,525 KB)
[v3]
Wed, 21 Oct 2026 09:05:50 UTC (28,262 KB)
来源：arXiv cs.AI · arxiv.org