In-House Subset of MFID for AI-Generated Face Detection

This repository provides the In-House Subset of MFID (Multidimensional Facial Image Dataset), a curated dataset designed for AI-generated face detection research. The dataset contains both authentic human face images and synthetic face images generated by multiple state-of-the-art generative models, including diffusion-based models such as Stable Diffusion 2/3 and Flux 1, as well as other modern generation pipelines. The dataset is constructed to support robust and generalizable deepfake detection by covering diverse demographic attributes (e.g., age, gender, pose) and multiple synthesis styles. It is intended as a benchmark for evaluating detection models under cross-model generalization scenarios. This dataset accompanies the paper: AI-Generated Face Detection Using Multi-Feature Fusion The proposed approach integrates multi-domain forensic features (spatial RGB signals, statistical texture descriptors, frequency spectrum anomalies, and structural edge cues) to detect subtle artifacts introduced by generative models. Experimental results in the paper demonstrate strong detection performance and generalization ability across heterogeneous generative architectures. The dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. Users are free to share and adapt the dataset for research purposes with proper attribution. File format: Image files organized by category (real vs. AI-generated) labels.csv provides binary labels for training and evaluation If you use this dataset, please cite our paper.本仓库提供MFID（Multidimensional Facial Image Dataset，多维面部图像数据集）的内部子集，这是一款专为AI生成人脸检测研究精心打造的精选数据集。该数据集同时包含真实人类面部图像，以及由多款当前顶尖生成模型生成的合成面部图像，涵盖基于扩散模型的Stable Diffusion 2/3、Flux 1等架构，以及其他现代生成流水线。 本数据集旨在通过覆盖多样化的人口统计学属性（如年龄、性别、姿态）与多种合成风格，支撑鲁棒且可泛化的深度伪造检测研究，其定位为用于评估跨模型泛化场景下检测模型性能的基准数据集。 本数据集配套发表的论文为《基于多特征融合的AI生成人脸检测》（AI-Generated Face Detection Using Multi-Feature Fusion）。该研究提出的方法整合了多域取证特征（空间RGB信号、统计纹理描述符、频谱异常与结构边缘线索），以识别生成模型引入的细微伪造痕迹。论文中的实验结果表明，该方法在异构生成架构下展现出优异的检测性能与泛化能力。 本数据集采用知识共享署名4.0国际许可（Creative Commons Attribution 4.0 International，CC BY 4.0）协议发布。使用者可自由分享并改编本数据集用于科研用途，但需注明正确的原作者署名。 文件格式： 图像文件按类别（真实人脸 vs AI生成人脸）进行组织。 labels.csv 文件提供用于训练与评估的二分类标签。 若您使用本数据集，请引用我们的论文。