Enhancing Neural Style Transfer with Gradient Loss for High-Quality Artistic Image Synthesis

Art is an incredible skill of humans and it is a great achievement of artificial intelligence that there exists an algorithm for artistic images, which was firstly developed by Leon A. Gatys. Gatys proposed an artificial system of fill up an image with desired styles. Although a lot of corrections have been made in this algorithm and some are effective, but cannot say that it has reached its full form and the research is still going on. This paper introduces optimal modifications to make the result of combined artistic style images with high perceptual quality. Here introduces a new loss term called the gradient loss, which identifies the edge features and makes it more precise for the high visual experience. As well as for getting the optimum loss value of style image, bring the smooth l1 loss instead of computing the mean squared error. The effectiveness of our model illustrates experiments.