Expression transfer

[davidvfx07]

Since 3D landmark output is possible, would it be possible to transfer the expressions from one tracked face to another while keeping the original head position (like this but with the landmarks)?

[1adrianb]

In principle it's possible, there a few systems around that use this landmark detector for this (deep face-like). However, using it directly its sub-optimal, as one will need to train a form of generative modeling, conditioned on the model output, to add the required visual realism.

[davidvfx07]

Are there any specific repos I can be referred to?

[Michele-Zhu]

Hi, if still needed this repo_link performs what you're asking not directly but with some knowledge of python & neural network you can easily make it work for photos. Takes some time to make it actually work if you're not too familiar with the tools used in the project