AIGC伪造检测数据集

面向中文场景下人脸视频防伪检测与社会风险防控需求，构建了一个大规模、多类型的人脸视频防伪基准数据集，涵盖真实人脸视频及面部替换、人脸重演、语音驱动和视听协同等多种 AIGC 伪造样本，贴近电信诈骗、身份冒用等真实攻击形态。数据通过“授权素材采集-伪造生成-审核标注”流程获取，并经自动化质量筛查与人工复核，完成伪造类型、模态特征、声画一致性等结构化标注及统一规范化处理，形成可追溯、可复现、可扩展的数据资源，可支撑多模态检测模型训练与规则构建。To address the requirements of face video anti-counterfeiting detection and social risk prevention and control in Chinese scenarios, we developed a large-scale, multi-type benchmark dataset for face video anti-counterfeiting. This dataset encompasses authentic face videos and various AIGC-generated counterfeit samples including face swapping, face reenactment, speech-driven forgery, and audio-visual collaborative forgery, which closely match real-world attack patterns such as telecom fraud and identity impersonation. The dataset is collected via the workflow of "authorized material collection - forgery generation - review and annotation", followed by automated quality screening and manual review. It has completed structured annotations covering forgery type, modal features, and audio-visual consistency, as well as unified standardized processing, thus forming a traceable, reproducible, and extensible data resource that supports the training of multimodal detection models and the establishment of detection rules.

搜集汇总

背景与挑战

背景概述

该数据集面向中文场景下人脸视频防伪检测与社会风险防控需求，构建了一个大规模、多类型的人脸视频防伪基准数据集，涵盖真实视频及多种AIGC伪造样本，贴近真实攻击形态。数据通过规范流程获取并经过质量筛查与标注处理，形成可追溯、可复现的资源，可用于支持多模态检测模型的训练与规则构建。