FaceForensics++ is a forensics dataset consisting of 1000 original video sequences that have been manipulated with four automated face manipulation methods: Deepfakes, Face2Face, FaceSwap and NeuralTextures. The data has been sourced from 977 youtube videos and all videos contain a trackable mostly frontal face without occlusions which enables automated tampering methods to generate realistic forgeries.

FaceForensics is a video dataset consisting of more than 500,000 frames containing faces from 1004 videos that can be used to study image or video forgeries. All videos are downloaded from Youtube and are cut down to short continuous clips that contain mostly frontal faces. This dataset has two versions:

Source-to-Target: where the authors reenact over 1000 videos with new facial expressions extracted from other videos, which e.g. can be used to train a classifier to detect fake images or videos.

Selfreenactment: where the authors use Face2Face to reenact the facial expressions of videos with their own facial expressions as input to get pairs of videos, which e.g. can be used to train supervised generative refinement models.

This is the FaceForensics++ dataset downloaded from original scripts. The dataset contains the following folders: DeepFakeDetection, Deepfakes, Face2Face, FaceShifter, FaceSwap, NeuralTextures, original, csv. Total 7010 files with 7000 mp4 videos (6000 deepfake, 1000 real) and 10 csv files

Folder Info:

DeepFakeDetection - 1000 deepfake videos.

Deepfakes - 1000 deepfake videos.

Face2Face - 1000 deepfake videos.

FaceShifter - 1000 deepfake videos.

FaceSwap - 1000 deepfake videos.

NeuralTextures - 1000 deepfake videos.

original - 1000 real videos.

csv - 10 csv files with metadata about the videos.

The videos are downloaded at moderate compression with c23 option ( c23 corresponds to H.264 compression at quality 23 (constant rate factor, CRF 23) )

More about FaceForensics++:

https://github.com/ondyari/FaceForensics

http://niessnerlab.org/projects/roessler2018faceforensics.html

LICENSE:

https://github.com/ondyari/FaceForensics/blob/master/LICENSE

Terms of Use:

https://kaldir.vc.in.tum.de/faceforensics/webpage/FaceForensics_TOS.pdf

With the advancement of deep forgery techniques, particularly propelled by generative adversarial networks (GANs), identifying deepfake faces has become increasingly challenging. Although existing forgery detection methods can identify tampering details within manipulated images, their effectiveness significantly diminishes in complex scenes, especially in low-quality images subjected to compression. To address this issue, we proposed a novel deep face forgery video detection model named Two-Stream Feature Domain Fusion Network (TSFF-Net). This model comprises spatial and frequency domain feature extraction branches, a feature extraction layer, and a Transformer layer. In the feature extraction module, we utilize the Scharr operator to extract edge features from facial images, while also integrating frequency domain information from these images. This combination enhances the model’s ability to detect low-quality deepfake videos. Experimental results demonstrate the superiority of our method, achieving detection accuracies of 97.7%, 91.0%, 98.9%, and 90.0% on the FaceForensics++ dataset for Deepfake, Face2Face, FaceSwap, and NeuralTextures forgeries, respectively. Additionally, our model exhibits promising results in cross-dataset experiments.. The code used in this study is available at: https://github.com/hwZHc/TSFF-Net.git.

Dataset

This dataset was created by Vittorio Stile

Released under Other (specified in description)

Contents

Dataset

This dataset was created by Siddhant Baiswar

Contents

With the advancement of deep forgery techniques, particularly propelled by generative adversarial networks (GANs), identifying deepfake faces has become increasingly challenging. Although existing forgery detection methods can identify tampering details within manipulated images, their effectiveness significantly diminishes in complex scenes, especially in low-quality images subjected to compression. To address this issue, we proposed a novel deep face forgery video detection model named Two-Stream Feature Domain Fusion Network (TSFF-Net). This model comprises spatial and frequency domain feature extraction branches, a feature extraction layer, and a Transformer layer. In the feature extraction module, we utilize the Scharr operator to extract edge features from facial images, while also integrating frequency domain information from these images. This combination enhances the model’s ability to detect low-quality deepfake videos. Experimental results demonstrate the superiority of our method, achieving detection accuracies of 97.7%, 91.0%, 98.9%, and 90.0% on the FaceForensics++ dataset for Deepfake, Face2Face, FaceSwap, and NeuralTextures forgeries, respectively. Additionally, our model exhibits promising results in cross-dataset experiments.. The code used in this study is available at: https://github.com/hwZHc/TSFF-Net.git.

Results of cross-database experiments on the Celeb-DF dataset (AUC).

Ablation study results for branching strategies (ACC).

such as FaceForensics++ and DFDC

As deepfakes pose increasing threats to media authenticity and public trust, accurate detection mechanisms are vital. This paper proposes a novel dual-stage transformer model for detecting deepfake facial videos. The model combines a spatial attention transformer to capture facial features across individual frames and a contextual consistency transformer to track identity coherence across time. Results from benchmark datasets (FaceForensics++, DFDC) show this architecture significantly outperforms conventional CNNs and RNNs in both accuracy and generalization. This research offers a pathway toward more robust and interpretable video forensics solutions.

In recent years, the abuse of a face swap technique called deepfake Deepfake has raised enormous public concerns. Deepfake manipulates deep learning techniques to replace one person's face in a video to someone else's without leaving obvious traces. So far, a large number of deepfake videos (also known as "deepfakes") have been crafted and uploaded to the internet, which calls for the development of effective countermeasures.One promising countermeasure against deepfakes is deepfake detection. Several deepfake datasets have been released to support the training and testing of deepfake detectors, such as DeepfakeDetection and FaceForensics++.While this has greatly advanced deepfake detection, most of the real videos in these datasets are filmed with a few volunteer actors in limited scenes, and the fake videos are crafted by researchers using a few popular deepfake softwares. Detectors developed on these datasets may lose effectiveness when applied to detect the vast variety of deepfake videos in the wild (those uploaded to varies video-sharing websites). To better support detection against real-world deepfakes, in this paper, we introduce a new dataset WildDeepfake, which consists of 7,314 face sequences extracted from 707 deepfake videos that are collected completely from the internet. WildDeepfake is a small dataset that can be used, in addition to existing datasets, to develop more effective detectors against real-world deepfakes.We conduct a systematic evaluation of a set of baseline detection networks on both existing and our WildDeepfake datasets, and show that WildDeepfake is indeed a more challenging dataset, where the detection performance can decrease drastically. We also propose two (eg. 2D and 3D) Attention-based Deepfake Detection Networks (ADDNets) to leverage the attention masks on real/fake faces for improved detection. We empirically verify the effectiveness of ADDNets on both existing and WildDeepfake.

The DFDC (Deepfake Detection Challenge) is a dataset for deepface detection consisting of more than 100,000 videos.

The DFDC dataset consists of two versions:

Preview dataset. with 5k videos. Featuring two facial modification algorithms. Full dataset, with 124k videos. Featuring eight facial modification algorithms

We build a hybrid fake face (HFF) dataset, which contains eight types of face images. For real face images, three types of face images are randomly selected from three open datasets. They are low-resolution face images from CelebA, high-resolution face images from CelebA-HQ, and face video frames from FaceForensics, respectively. Thus, real face images under internet scenarios are simulated as real as possible. Then, some most representative face manipulation techniques, which include PGGAN and StyleGAN for identity manipulation, Face2Face and Glow for face expression manipulation, and StarGAN for face attribute transfer, are selected to produce fake face images. The HFF dataset is a large fake face dataset, which contains more than 155k face images.