Bidimensional face morph

In face rehabilitation, patients are asked to perform a face deformation and see how well they can deform their faces. This project aims to give a way to measure this deformation objectively.

Our approach uses frames taken from a colored video to create a 3D model of the person's relaxed face. Subsequently, a second model is created starting from the frames of the video where a certain grimace is performed and through the comparison of the models the extent and intensity with which the grimace was performed is detected.

Requirements

The following are the Python libraries required:

• numpy
• matplotlib
• h5py
• face_alignment

You also need to download the following repositories and rename their directories in the specified way:

• deep-3dmm-refinement as deep_3dmm_refinement
• SLC-3DMM as SLC_3DMM

Datasets

Download data3dmm for the average model components and put it inside a directory called data3dmm.

You must put your frames in a directory and separate them in subdirectories called 0001, 0002, etc. Each of these subdirectories contains the frames of a video of someone's face that starts from a relaxed facial expression and end up with some kind of grimace. The frames must be jpg images called as sequential numbers (e. g. 1916.jpg, 1917.jpg, ...)

Usage

Run the main.py file passing to it the directory in which you have all of the frames (e. g. python3 main.py LR/claudio_ferrari/rgbReg_frames).

This will create and populate the following output directories:

• landmarks_2d where the images with the 2D landmarks will be saved;
• deformation_heatmaps where your heatmaps for the face deformation will end up.