arXiv 2602.172602026-03-05cs.CV

EA-Swin：一种嵌入无关的Swin Transformer用于AI生成视频检测

EA-Swin: An Embedding-Agnostic Swin Transformer for AI-Generated Video Detection

N2TP Technology Solution JSC（N2TP技术解决方案有限公司）；College of Technology, National Economics University（国家经济大学技术学院）；University of Science, Vietnam National University（越南国家大学科学大学）；Faculty of AI and Data Science, Phenikaa University（Phenikaa大学人工智能与数据科学学院）；The Saigon International University（西贡国际大学）；College of Informatics, Yuan Ze University（元智大学资讯学院）

Hung Mai, Loi Dinh, Duc Hai Nguyen, Dat Do, Luong Doan, Khanh Nguyen Quoc, Huan Vu, Naeem Ul Islam, Tuan Do

备注：: 2nd preprint version

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

EA-Swin通过因子化窗口注意力设计，直接建模预训练视频嵌入的时空依赖性，实现对AI生成视频的高准确率检测，优于现有方法5-20%。

摘要：

近年来，基础视频生成器如Sora2、Veo3及其他商业系统已产生高度逼真的合成视频，暴露了现有检测方法的局限性，这些方法依赖于浅层嵌入轨迹、基于图像的适应或计算密集型MLLMs。我们提出EA-Swin，一种嵌入无关的Swin Transformer，通过因子化窗口注意力设计直接在预训练视频嵌入上建模时空依赖性，使其兼容通用ViT风格的基于补丁的编码器。此外，我们构建了EA-Video数据集，一个包含13万视频的基准数据集，整合了新收集的样本与精心挑选的现有数据集，涵盖多样化的商业和开源生成器，并包含未见生成器分割以进行严格跨分布评估。广泛的实验表明，EA-Swin在主要生成器上实现了0.97-0.99的准确率，比先前最先进的方法（通常为0.8-0.9）高出5-20%，同时在未见分布上保持强大的泛化能力，建立了现代AI生成视频检测的可扩展且稳健的解决方案。

英文摘要：

Recent advances in foundation video generators such as Sora2, Veo3, and other commercial systems have produced highly realistic synthetic videos, exposing the limitations of existing detection methods that rely on shallow embedding trajectories, image-based adaptation, or computationally heavy MLLMs. We propose EA-Swin, an Embedding-Agnostic Swin Transformer that models spatiotemporal dependencies directly on pretrained video embeddings via a factorized windowed attention design, making it compatible with generic ViT-style patch-based encoders. Moreover, we construct the EA-Video dataset, a benchmark dataset comprising 130K videos that integrates newly collected samples with curated existing datasets, covering diverse commercial and open-source generators and including unseen-generator splits for rigorous cross-distribution evaluation. Extensive experiments show that EA-Swin achieves 0.97-0.99 accuracy across major generators, outperforming prior SoTA methods (typically 0.8-0.9) by a margin of 5-20\%, while maintaining strong generalization to unseen distributions, establishing a scalable and robust solution for modern AI-generated video detection.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。