Generování nočních obrazů z denních

Abstract

This paper deals with the topic of generative adversarial networks and their use for the generation of the nigh-time images from the day-time ones. Initially, the theoretical part introduces the architectures, techniques and properties of deep neural networks, which are later applied. Next, the preparation and characterization of the training and test dataset from the provided materials is described. Then, the paper describes the implementation of the appropriate methods and architectures selected based on the theoretical analysis and details the experiments. The final section compares the results of the different experiments and evaluates them.