ForenSynths

该数据集名为ForenSynths，包含真实图像以及基于13种不同的基于卷积神经网络（CNN）的生成对抗网络（GAN）模型生成的DeepFake图像，主要关注于人脸图像。在数据划分上，训练集仅包含由ProGAN模型生成的DeepFake图像，而测试集则包含了来自StarGAN和WhichFaceIsReal等其他模型生成的图像。具体规模上，该数据集包含了720119张训练图像，8000张验证图像以及90310张测试图像。该数据集的任务是进行Deepfake检测。

The dataset is named ForenSynths. It comprises real facial images as well as DeepFake images generated by 13 distinct Convolutional Neural Network (CNN)-based Generative Adversarial Network (GAN) models, with a primary focus on facial imagery. Regarding data partitioning, the training set exclusively includes DeepFake images generated by the ProGAN model, while the test set contains images generated by other models such as StarGAN and WhichFaceIsReal. Specifically, the dataset consists of 720,119 training images, 8,000 validation images, and 90,310 test images. This dataset is designed for deepfake detection tasks.

搜集汇总

背景与挑战

背景概述

ForenSynths数据集是CNNDetection项目的一部分，包含多种CNN生成图像的真实与合成样本，用于训练和评估检测模型。数据集结构清晰，分为训练集、测试集和验证集，支持多种CNN生成算法的检测任务。