This is a parent repository containing a few other repositories used to adapt and test models for Farapy V2. The descriptions for each submodule are provided below.
This contains the implementation of the final model used for Farapy V2 (MAE-Face). It uses a masked autoencoder pretraining phase where it is trained on a large general dataset of face images. It is then fine-tuned on the downstream task of FAU intensity estimation.
This repository contains some data/label processing tools for use with the DISFA and FEAFA+ datasets. The tools allow for generating subject-independent train/test folds following the method of similar FAU recognition/intensity estimation works.
This repository contains an adapted version of ME-GraphAU for FAU unilateral intensity estimation. It was our initial approach, but ultimately did not provide adequate performance.
This repository contains the Pytorch implementation of Farapy V1's teacher model.