De-Aging with Deepfakes - BSc Dissertation
The term "Deepfake” refers to a digitally altered video which replaces one actor’s face with another’s. This is accomplished through machine learning techniques which train a model based on the reference footage that the user provides
My dissertation project researched whether the use of deepfake technology to de-age an actor is effective, both in terms of its’ convincingness and the time it takes to achieve the effect. This was compared against CG methods such as those used in Gemini Man (2019) through perception tests with participants.
The results showed that the method was successful in convincing participants that the actor was the desired age, however the majority of participants were able to tell that the actor had been digitally altered. This issue could be solved with a variety of techniques, such as manual compositing, training the model to a higher number of iterations, or using more training data.
My full dissertation is available below, and a playlist of the clips used within the perception tests can be found here.
