Train a multi-task learning model to estimate age, gender and emotion of an input face. The trianing is done from multiple datasets, with a mixed-batch strategy. The code is written in Keras with Tensorflow as the backend.
The labels for age is from LAP dataset. The labels for gender and emotion is from MTFL dataset.
A similar implementation in Tensorflow is here: https://github.com/li-js/Face_age_gender_emotion