This work is part of a new national collaborative project in France ANR ASTRID Guerre Cognitive, proposed by EURECOM specialized in computer vision and IRCAM specialized in audio, entitled: “Fight against deepfakes of French personalities”.
Recent challenges have shown that it is extremely difficult to develop universal detectors for deepfake videos - such as the deepfakes used to forge a person's identity. When the detectors are exposed to videos generated by a new algorithm, i.e. unseen during the training phase, the performance remains limited. For the video part, the algorithms check frames one by one, without considering facial dynamics. This is a major weakness of deepfake video generators. The present project aims at implementing and training customized deepfake detection algorithms on individuals for which many real and fake audio-video sequences are available and/or can be created. Based on state-of-the-art audio and video algorithms, the thesis will focus on considering the temporal evolution of audio-visual signals and their synchronization in the generation and detection of deepfakes. The objective is to demonstrate that by using audio and video simultaneously and focusing on a specific person during training and detection, it is possible to design efficient detectors even against unseen generators.
- Education Level / Degree: PhD
- Field / specialty: Image processing / Computer Vision / Artificial Intelligence
The application must include:
- Detailed curriculum,
- Name and address of 2 references.
Applications should be submitted by e-mail to firstname.lastname@example.org with the reference:
Start date: ASAP
Duration: 18-months CDD (temporary work contract)