A Review on Neural Style Transfer with Auto Text Generation

Abstract

Neural Style Transfer has been a topic of discussion since the past few years, using which any simple content image can be converted in the style of a particular artist. Several methods have been proposed so far for the implementation of Style Transfer, and we discuss a few of these approaches in this paper, (semantic style transfer and photorealistic style transfer) and compare them accordingly. Text generation using neural networks is another field of interest, which has been explored by various researchers in the past. This paper further discusses various text generation models created in the past; using character based Long-Short term memory (LSTM), Recurrent Neural Network Language Model (RNNLM) as well as Multi Adversarial Training. Both of these technologies (neural style transfer and auto text generation) show potential to further be combined, such that the text is generated in the context of the stylized image.