Cross-Modal Deepfake Dataset (CMDFD)

该数据集是一个全面的集合，涵盖了用于评估多种跨模态深度伪造检测方法的跨模态深度伪造生成技术。它包含了使用Wav2Lip、VideoReTalking、Audio2Head和MakeItTalk等多种操纵技术生成的伪造内容。每个身份都经过了另一个人的语音以及原始音频的操纵。该数据集的规模为：从1,000个身份中生成了8,000个假视频，使用了四种伪造方法。其任务是进行深度伪造检测。

This dataset is a comprehensive collection encompassing cross-modal deepfake generation techniques intended for evaluating various cross-modal deepfake detection methods. It contains forged content generated by multiple manipulation technologies including Wav2Lip, VideoReTalking, Audio2Head and MakeItTalk. For each identity, manipulation is performed using both its original audio and the voice of another individual. In terms of scale, 8,000 fake videos are generated from 1,000 distinct identities using four forgery methods. The core task supported by this dataset is deepfake detection.

搜集汇总

背景与挑战

背景概述

Cross-Modal Deepfake Dataset (CMDFD) 是一个用于评估跨模态深度伪造检测方法的综合数据集，包含使用Wav2Lip、VideoReTalking、Audio2Head和MakeItTalk等多种技术生成的伪造内容，涉及语音和音频操纵。数据集规模为从1,000个身份生成8,000个假视频，旨在支持深度伪造检测任务的研究和开发。