Image / Video DeepFake Detection service

The present DeepFake detection service receives the URL address of an image or video as input and generates a single DeepFake probability score as output. The input triggers a multi-stage processing pipeline, including dedicated functions for the downloading and pre-processing of the input for the extraction of the contained faces. The detected faces are submitted to an ensemble scheme of five DeepFake detection models. The outputs are aggregated to derive a single probability score indicating whether the input medium contains DeepFake faces.

As a Service

Link to the service (might require credentials)

Link to the service's model card

Link to our research team's website.

Developed by

Centre for Research & Technology, Hellas - CERTH

Main Characteristic

Supports both images and video from a plethora of social media and file sharing websites,
Handles multiple faces via face-recognition based clustering described in [1],
Aggregates predictions from multiple persons-of-interest focusing on confident final scores.

For a detailed description of the service please see the associated paper [2].

[1] https://arxiv.org/abs/2006.07084

[2] https://arxiv.org/abs/2204.12816

Technical Categories

AI services Computer vision Machine learning

Keywords