Deep Fakes Dataset (inamibora)

Deep Fakes数据集是用于deepfake检测的 “野外” 肖像视频的集合。数据集中的视频在源生成模型，分辨率，压缩，照明，纵横比，帧速率，运动，姿势，化妆品，遮挡，内容和上下文方面都是不同的真实样本。它们来自各种来源，例如新闻文章，论坛，应用程序和研究演示; 总计多达142个视频，32分钟和17 gb。在可能的情况下，合成视频将与原始视频匹配。

The Deep Fakes Dataset is a collection of "in-the-wild" portrait videos for deepfake detection. The videos within the dataset are authentic samples that exhibit diversity across a range of dimensions, including source generation models, resolution, compression, lighting, aspect ratio, frame rate, motion, pose, cosmetics, occlusions, content, and context. These videos are sourced from various platforms such as news articles, forums, applications, and research demonstrations, totaling up to 142 videos, 32 minutes of footage, and 17 GB of data. Where feasible, synthetic videos will be paired with their corresponding original authentic videos.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于深度伪造检测的肖像视频集合，包含142个视频，总计32分钟和17GB，视频在生成模型、分辨率等方面具有多样性。它由宾汉顿大学和Intel于2019年发布，适用于研究目的。