arXiv cs.AI· Zheyuan Gu, Minghao Shao, Zhen Wang, Keyu Mao, Ailiang Lin, Mingyou Liang, Mingkun Xu, Shijie Zhang, Hao Jiang, Yusong Wang·· 4 小时前AI 评分37
SpInShield:揭示并缓解深度伪造视频检测中的时序攻击
Exposing and Mitigating Temporal Attack in Deepfake Video Detection
AI 导读
arXiv 论文指出时空深度伪造检测器虽 AUC 很高,但易过拟合脆弱的时序频谱线索而遭规避攻击。作者提出 SpInShield,一种时序频谱不变的防御框架,用可学习频谱对抗合成极端频谱形变,并以 shortcut 抑制优化迫使编码器提取可靠取证线索。实验显示其在模拟幅度频谱攻击下 AUC 超最强基线 21.30 个百分点,在常用数据集上表现具竞争力。
正文 · 原文
Abstract:While spatiotemporal deepfake detectors achieve high AUC, our experiments reveal their susceptibility to evasion attacks. These models tend to overfit on fragile temporal spectrum cues, rather than learning robust semantic causality. To mitigate this vulnerability, we propose SpInShield, a temporal spectral-invariant defense framework explicitly designed to decouple semantic motion from manipulatable spectral artifacts. We propose a learnable spectral adversary that dynamically synthesizes severe spectral deformations, simulating extreme attack scenarios. By employing a shortcut suppression optimization strategy, SpInShield compels the encoder to extract reliable forensic cues while purging unstable spectral statistics from the latent space. Experiments show that SpInShield obtains competitive performance on widely used datasets and outperforms the strongest baseline by 21.30 percentage points in AUC under simulated amplitude spectral attacks.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.07398 [cs.CV] |
| (or arXiv:2605.07398v2 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2605.07398 arXiv-issued DOI via DataCite |
Submission history
From: Zheyuan Gu [view email]
[v1]
Fri, 8 May 2026 07:53:57 UTC (5,467 KB)
[v2]
Fri, 13 Nov 2026 18:41:47 UTC (5,463 KB)
来源:arXiv cs.AI · arxiv.org