Machine learning aplicat a la generació de moodboards

Abstract

[en] Moodboards are used to represent objects that are formed by the composition of other objects and they try to obtain a result of cohesion among them and try to generate a new image. Machine learning allows us to train a machine so that it can generate the values with the data that we have entered previously, which must be correct data. What we want to obtain is a combination between the two parts to achieve the generation of new images, with correct and cohesive objects and different from the images used to train the machine. So the objective of this project is to create a machine that allows us to create moodboards and generate correct it without the need of a person to create them, and design an automated system to replace a slowly non automated task. Apart from creating this system, we will also analyze it’s efficiency, between the non-automated system and the automated system, so we can establish whether the system created is efficient enough, fast and accurate generating moodboards. Finally, these tests will be carried out by supervising expert people in order to improve the evaluation.