Task I: Deepfake detection task. Given a set of Real and Deepfake images created by different GAN engines, the objective is to create a detector able to correctly classify Deepfake images in any scenario.
Task II: Source image reconstruction task.
ABSTRACT
Multimedia data manipulation and forgery has never been easier than today, thanks to the power of Artificial Intelligence (AI). AI-generated fake content, commonly called Deepfakes, have been raising new issues and concerns, but also new challenges for the research community. The Deepfake detection task has become widely addressed, but unfortunately, approaches in the literature suffer from generalization issues. In this paper, the Face Deepfake Detection and Reconstruction Challenge is described. Two different tasks were proposed to the participants: (i) creating a Deepfake detector capable of working in an “in the wild” scenario; (ii) creating a method capable of reconstructing original images from Deepfakes. Real images from CelebA and FFHQ and Deepfake images created by StarGAN, StarGAN-v2, StyleGAN, StyleGAN2, AttGAN and GDWCT were collected for the competition. The winning teams were chosen with respect to the highest classification accuracy value (Task I) and “minimum average distance to Manhattan” (Task II). Deep Learning algorithms, particularly those based on the EfficientNet architecture, achieved the best results in Task I. No winners were proclaimed for Task II. A detailed discussion of teams’ proposed methods with corresponding ranking is presented in this paper.
Download Paper |
Different Deepfake images were generated for the competition [WEB-PAGE].
As regards TASK 1: DEEPFAKE DETECTION TASK, two datasets of real face images were used for the employed experimental phase: CelebA and FFHQ. Different Deepfake images were generated
considering StarGAN, GDWCT, AttGAN, StyleGAN and StyleGAN2 architectures. For TASK 2: SOURCE IMAGE RECONSTRUCTION, deepfake images were created vy using StarGAN-v2 architecture.
Download Labels (each row of the label file contains the name of the image followed by the label)
Manipulation details
(each row contains the name of the image followed by a list of manipulations applied to the image. If the list is empty, then the image is not subject to attack, otherwise the manipulation type and its parameters are specified (e.g., JPEG compr: 84 represents JPEG compression with quality factor equal to 84; Rotation: 225 degrees represents applied rotation filter with degree value equal to 225; Gaussian: 9 represents Gaussian blur filter with kernel size equal to 9; etc.). For more details on the manipulation operations applied to images, please read the official paper.
Download Labels (each row of the label file contains the name of the deepfake image followed by the name of the source image)
Source images are available here [DOWNLOAD FOLDER]
REFERENCES