Hello, my name is Ruben Jentschke. I’m a student at Stuttgart Media University, and I’m writing my thesis in the Unreal Engine and Metahuman field.
I’m currently researching the use of Metahuman to create a 2D dataset to train 2D face-tracking algorithms and additionally provide 3D position data for the algorithm of the tracker. The advantage of creating training data with Metahuman is that it is based on 3D models and could, therefore, serve as a ground truth compared to, e.g., 2D animations where you would place 3D landmarks on the 2D footage manually. (With the risk of inaccuracies and the lack of scalability)
I’ve already made some attempts in Unreal Engine to see if absolute points can be accessed on the Metahuman mesh independent from individual Metahuman characters. I’m looking for fixed positions like eyelids, eyes, ears, etc. The ideal would be to read out position data from the animation rig or the “bone” structure to avoid manually annotating the meshes.
It would be really helpful if you could help me find a starting point or forward me to other helpful contacts.
Thanks for your support in advance.