Multiple Facial Attributes Estimation Based on Weighted Heterogeneous Learning

To estimate multiple face attributes, independent classifier for each attribute are trained such as facial point detection, gender recognition, and age estimation in the conventional approach. It is inefficient because the computational cost of training and testing increases with the number of tasks. To address this problem, heterogeneous learning is able to train a single classifier to perform multiple tasks. Heterogeneous learning is simultaneously train regression and recognition tasks, thereby reducing both training and testing time. However, it is difficult to obtain equivalent performance for set of single task classifiers due to variance of training error of each task. In this paper, we propose weighted heterogeneous learning of a convolutional neural network with a weighted error function. Our method outperformed the conventional method in terms of facial attribute recognition, especially for regression tasks such as facial point detection, age estimation, and smile ratio estimation.