The convolutional neural network (CNN) works very well in many computer vision tasks including the face-related problems. However, in the case of age estimation and facial expression recognition (FER), the accuracy provided by the CNN is still not good enough to be used for the real-world problems. It seems that the CNN does not well find the subtle differences in thickness and amount of wrinkles on the face,which are the essential features for the age estimation and FER. Also, the face images in the real world have many variations due to the face rotation and illumination, where the CNN is not robust in finding the rotated objects when not every possible variation is in the training data. Moreover, The Multi Task Learning (MTL) Based based methods can be much helpful to achieve the real-time visual understanding of a dynamic scene, as they are able to perform several different perceptual tasks simultaneously and efficiently. In the exemplary MTL methods, we need to consider constructing a dataset that contains all the labels for different tasks together. However, as the target task becomes multi-faceted and more complicated, sometimes unduly large dataset with stronger labels is required. Hence, the cost of generating desired labeled data for complicated learning tasks is often an obstacle, especially for multi-task learning. Therefore, first to alleviate these problems, we first propose few methods in order to improve single task baseline performance using gabor filters and Capsule Based Networks , Then We propose a new semi-supervised learning method on face-related tasks based on Multi-Task Learning (MTL) and data distillation.
【 预 览 】
附件列表
Files
Size
Format
View
Deep Multi-Dataset Multi-Domain Multi-Task Frame Works for Facial Expression Recognition, Age and Gender Estimation