The Deepfake Detection Challenge (DFDC) Preview Dataset

View on arXiv ← Back to list

Authors: Brian Dolhansky, Russ Howes, Ben Pflaum, Nicole Baram, Cristian Canton Ferrer

Published: 2019-10-19 22:35:52+00:00

AI Summary

This paper introduces a preview of the Deepfakes Detection Challenge (DFDC) dataset, comprising 5,000 videos generated using two facial modification algorithms. The dataset prioritizes diversity in actor demographics and recording conditions, and defines novel metrics for evaluation, also providing baseline performance results from existing deepfake detection models.

Abstract

In this paper, we introduce a preview of the Deepfakes Detection Challenge (DFDC) dataset consisting of 5K videos featuring two facial modification algorithms. A data collection campaign has been carried out where participating actors have entered into an agreement to the use and manipulation of their likenesses in our creation of the dataset. Diversity in several axes (gender, skin-tone, age, etc.) has been considered and actors recorded videos with arbitrary backgrounds thus bringing visual variability. Finally, a set of specific metrics to evaluate the performance have been defined and two existing models for detecting deepfakes have been tested to provide a reference performance baseline. The DFDC dataset preview can be downloaded at: deepfakedetectionchallenge.ai

Key findings

The authors present a new dataset for deepfake detection focusing on diversity and actor consent. Baseline results using existing models demonstrate the challenge of robust deepfake detection, highlighting the need for more advanced techniques. Novel evaluation metrics that account for the prevalence of deepfakes in real-world scenarios were introduced.

Approach

The authors created a dataset of videos with diverse actors, backgrounds, and lighting conditions. Two deepfake generation methods were used to produce the manipulated videos. Existing deepfake detection models (TamperNet and XceptionNet) were used to establish a baseline performance.

Datasets

DFDC preview dataset (5000 videos), FaceForensics dataset (for training XceptionNet)

Model(s)

TamperNet, XceptionNet (face and full-image versions)

Author countries

UNKNOWN

← Previous