DeepfakeTIMIT 深度伪造检测数据集
收藏资源简介:
DeepfakeTIMIT 是由 Idiap 研究所于 2018 年创建的一个用于深度伪造检测的数据集,它包含了使用开源的基于生成对抗网络 (GAN) 的方法交换面孔的视频。这些视频基于原始的自动编码器基础的 Deepfake 算法创建。数据集从公开可用的 VidTIMIT 数据库中手动选择了 16 对人脸相似的个体,为每个个体训练了两种不同质量的模型:低质量 (LQ) 模型输入 / 输出尺寸为 64×64,以及高质量 (HQ) 模型,尺寸为 128×128 。对于 VidTIMIT 数据库中每个人 10 个视频,生成了 320 个相应版本的视频,总共产生了 620 个面孔被交换的视频。音频方面,保留了每个视频的原始音轨,没有对音频通道进行任何修改。
DeepfakeTIMIT is a deepfake detection dataset created by the Idiap Research Institute in 2018. It contains videos of face swapping generated using open-source generative adversarial network (GAN)-based methods. These videos are developed based on the original autoencoder-based Deepfake algorithm. The dataset manually selected 16 pairs of individuals with visually similar facial features from the publicly available VidTIMIT database. Two models with different quality levels were trained for each individual: the low-quality (LQ) model with an input/output size of 64×64, and the high-quality (HQ) model with a size of 128×128. For the 10 videos per individual in the VidTIMIT database, 320 corresponding modified videos were generated, resulting in a total of 620 face-swapped videos. For the audio part, the original audio track of each video was retained without any modifications made to the audio channels.




