International Deepfake Detection Contest
The Global Fact-Checking Network (GFCN) invites the creators of deepfake detection systems to participate in the Deepfake Detection Contest and test the effectiveness of their software. Periodic quality analysis allows not only to evaluate the accuracy of each model, but also contributes to the discovery of new approaches in countering deepfakes.
The results of the contest will be presented in the fall.
Regulations
To test your model in the competition, you must use a video from the proposed asset. The asset is a balanced set of data, containing 30 video files, which are divided based on the following types of deepfake attacks:
- Face swap
- Lip sync
- Real videos without any signs of being a deepfake
The videos have a resolution of at least 480p, and to increase the objectivity of testing, these videos come from a variety of sources and depict different people.
How to participate
The process of participation in the contest consists of several key stages:
Step 01
Familiarization with the methodology: Before starting the testing process, it is recommended that you familiarize yourself with the verification process and its criteria. You can do this by clicking on the “Testing Methodology” button below.
Step 02
Downloading the dataset: To test the models, you will need to download a suitable set of videos. You can do this by using the “Download the Asset” button provided below.
Step 03
Results: After completing the testing, the results should be sent to the following email address info@gfcn.media
The results file should contain tables with video IDs and estimated values based on model predictions. To further analyze the model’s performance, additional data from metrics such as Precision, Recall, and Accuracy can be included.
Step 04
Based on the results of the testing, the GFCN will invite participants with the best results to a more in-depth discussion of methods and approaches, either in a face-to-face or online setting. The GFCN will also offer to validate results, exchange datasets and experiences in detecting deepfake materials. For more information on the GFCN’s events, click here.