[论文] ARMOR++: Agentic Orchestration of a Multi-Domain Primitive Set fo...

论文概要

研究领域: CV 作者: Christos Korgialas, Gabriel Lee Jun Rong, Dion Jia Xu Ho 发布时间: 2025-07-16 arXiv: 2507.12500

中文摘要

深度伪造检测器在黑盒对抗迁移场景下的可靠性频繁下降，因为这些模型往往依赖脆弱且架构相关的法医线索。现有迁移攻击通常缺乏语义感知能力，在严格无查询约束下难以保持有效性，尤其是当扰动从卷积代理模型迁移到基于Transformer的目标模型时。为应对这些局限，本文提出 ARMOR++——一个专为高可迁移性深度伪造逃逸设计的稳健多智能体框架。该框架利用 Qwen2.5-VL 视觉语言模型（VLM）提供空间语义先验，同时由 Qwen3 大语言模型（LLM）编排基元选择、自适应超参数重参数化和熵正则化扰动混合。通过整合五种互补基元——涵盖稠密优化、显著性方法、空间变换、频域扰动和块结构修改——ARMOR++ 有效针对异构归纳偏见。在 AADD-2025 基准上的严格评估表明，ARMOR++ 在低质量与高质量图像场景下均显著优于现有智能体和非智能体基线。统计分析证实，相比最先进的智能体基线，盲目标攻击成功率（ASR）获得大幅提升，且在面对非智能体基准和稳健防御配置时均表现出进一步优势。这些发现揭示了当前深度伪造检测器部署中存在显著的残余可靠性缺口，并证明了智能体编排在发现潜在漏洞方面的有效性。

原文摘要

The reliability of deepfake detectors frequently degrades under black-box adversarial transfer, as these models often rely on fragile, architecture-dependent forensic cues. Existing transfer attacks often lack semantic awareness and struggle to maintain effectiveness under strict no-query constraints, particularly when perturbations are transferred from convolutional surrogates to transformer-based targets. To address these limitations, this paper introduces ARMOR++, a robust multi-agent framework designed for high-transferability deepfake evasion. The framework leverages the Qwen2.5-VL Vision-Language Model (VLM) to supply spatial semantic priors, while the Qwen3 Large Language Model (LLM) orchestrates primitive selection, adaptive hyperparameter reparameterization, and entropy-regularized perturbation mixing. By integrating five complementary primitives, spanning dense optimization, sa...

--- *自动采集于 2026-07-19*

#论文 #arXiv #CV #小凯