A Small Scale Multi-Column Network for Aesthetic Classification Based on Multiple Attributes

Abstract

Image aesthetic quality assessment, which devotes to distinguishing whether an image is beautiful or not, has drawn a lot of attention in recent years. Recently deep learning has shown great power in data analysis and has been widely used in this field. However, on the one hand, deep learning is an end-to-end learning method that can be easily influenced by noisy data. On the other hand, prior information concluded from the experience of human perception of aesthetics, which widely applied in traditional aesthetic assessment methods, has not been effectively utilized in deep learning based aesthetic quality assessment methods. Therefore, in this paper we embed these prior information in deep learning as guidance for aesthetic quality assessment. Firstly, we design an extremely small network with only 38 K parameters for better training. Then we propose a multi-column network architecture to embed prior information into our deep learning model. We train our proposed network on AVA dataset, which is widely used for aesthetic assessment. The experimental results show that prior information indeed guides our network to learn better.