Deep Learning for Neural Style Transfer

Abstract:
The neural style transfer (NST) is an important task in the computer vision area. It aims to combine the content in an image with other images and generate an image with a specific style. This technique has been applied to many downstream image-related tasks and applications. Recently, the NST has made great progress with Convolution Neural Network (CNN), Generative Adversarial Network (GAN), and their variants. Thus, it is necessary to make a review of current work in NST. In this review, we firstly introduce the definition of NST and different types of NST algorithms. Then, we demonstrate some commonly used datasets and some indicators to evaluate the performance of different NST models. Finally, we analyze related work of NST and summarize some potential challenges and remained problems of state-of-the-art (SOTA) researches. This review will not only provide a good reference for researchers but also provide guidelines for some novices.