Style transfer with convolutional neural networks
Moreno Ramírez, Luís Fernando · 2018
The main objective of this thesis is to analyze an algorithm known as Neural Style Transfer which consists in transfer artistic styles from renowned painters as Vincent Van Gogh, Claude Monet, among others painters to any picture. For this purpose, I divide this work in three chapters to explain important topics to understand Neural Style Transfer. In chapter 1, I review the prerequisites needed for the algorithm from linear algebra, probability, vector calculus and matrix calculus. In chapter 2, In order to familiarize the reader with the neural network notion, I explain concepts from machine learning like supervised learning, regression, classification, underfitting and overfitting and the gradient descent algorithm. In chapter 3, I analyze the classification problem through linear classification by least squares and perceptron, and nonlinear classification through neural networks, which is a machine learning algorithm used today to make pattern recognition and autonomous driving. To finally explain, how the style transfer algorithm works.