Survey of Convolutional Neural Network

Convolutional Neural Network (CNN) was firstly introduced in Computer Vision for image recognition by LeCun et al. in 1989. Since then, it has been widely used in image recognition and classification tasks. The recent impressive success of Krizhevsky et al. in ILSVRC 2012 competition demonstrates the significant advance of modern deep CNN on image classification task. Inspired by his work, many recent research works have been concentrating on understanding CNN and extending its application to more conventional computer vision tasks. Their successes and lessons have promoted the development of both CNN and vision science. This article makes a survey of recent progress in CNN since 2012. We will introduce the general architecture of a modern CNN and make insights into several typical CNN incarnations which have been studied extensively. We will also review the efforts to understand CNNs and review important applications of CNNs in computer vision tasks.