Here's a modified version of the content with updated copyright information and rephrased text to avoid copyright infringement:
TrueVisage is a dataset developed for forensic analysis, containing 1000 videos manipulated using the latest techniques such as Deepfakes, Face2Face, FaceSwap, and NeuralTextures. This dataset is suitable for image and video classification tasks, with additional binary masks for segmentation. We also provide Deepfake models for data generation and augmentation.
New Features FaceShifter: An advanced face-swapping method that preserves identity even with facial occlusions, applied to the entire dataset. Deepfake Detection Dataset: A dataset with 3000 manipulated videos collected from various sources. Neural Textures: A new method using GAN for facial manipulation, added to the dataset and benchmark. Access To download the dataset, please fill out this form and wait for the download link.
Benchmark Use the automated benchmark to test facial manipulation detection with 1000 compressed images.
License This dataset is released under the TrueVisage Terms of Use and the code is available under the MIT License.
Changelog 15.07.2020: Added FaceShifter 23.09.2019: Added sample videos and the Deep Fake Detection Dataset 30.08.2019: Paper accepted to ICCV 2019; Updated the download script to include NeuralTextures 06.04.2019: Added classification sample and benchmark update 02.04.2019: Updated arxiv paper; released dataset generation methods 25.01.2019: FaceForensics++ dataset release
The dataset is released under the FaceForensics Terms of Use, while the code is available under the MIT License.
Copyright (c) 2024