My Diploma thesis for Auth was about using models, based on neural networks, for extracting data and parameters of facial expressions from images and video. Our goal was to create 3D facial reconstructions depicting human expressions' instances or animations. These were used to create a dataset for use in tasks related to facial expression recognition.
DECA (Detailed Expression Capture and Animation) model/software was used for extracting only facial parameters. The method creates 3D reconstructions of a human head with the expression of input images or videos. Moreover, it extracts the skin texture of the image and uses it in the 3D reconstruction. Then using a Python script, we edit and optimize the automatically produced textures.
Finally, the data were used in simulations to create a dataset with images of human expressions obtained with various capture parameters.
Example images of the results