Convolutional neural networks (CNN) have been a hot topic in the computer vision field. This paper attempts to use CNN for age estimation.

One contribution of the paper is the IMDB-WIKI dataset, which can be used to crawl images of celebrities from IMDb (http://www.imdb.com) and Wikipedia (https://en.wikipedia.org/).

For age estimation, the authors propose a deep expectation (DEX) approach. First, the images go through a face alignment process, which involves running the images through a detector for faces and rotating them. This is to form the 256-by-256-pixel face-aligned image for the input for the next step. Next, the DEX employs the CNN with the VGG-16 architecture to predict the age of a person. In this step, the CNN is trained on facial images with known age, and then takes an aligned face with context as input and returns the prediction of the age. Extensive experiments are conducted to evaluate the proposed method compared to other methods. The proposed model provides a better method for age estimation.

Aside from the IMDB-WIKI dataset, the main contribution of this paper is the presentation of the DEX system for age estimation. The proposed model helps to provide a better method for age estimation.

Future work is discussed in the paper as well. I would like to see more discussion about why the VGG-16 architecture is used, as opposed to VGG-19 or another. I’m sure if the DEX uses different architectures, it will get different results. It would be interesting to see the comparison of different architectures for age estimation.