Image Generation Based on Artworks Classified by Emotional Categories

Recent developments in the field of artificial neural networks have allowed researchers to create models that can generate artificial images of different content or modify existing images according to some pattern, for example, place bodies in different poses, rotate objects. However, not all tasks in this area have already been solved. The goal of this project is to create abstract images that evoke positive or negative emotions in the observer. Generative adversarial networks were chosen as a method. As a result of the project, we developed two models that generate abstract images evoking positive and negative emotions in the audience. The website and survey were created to evaluate the results. The models and generated images can be used for both scientific and business purposes. We also developed an architecture to enrich the data and improve the metrics of the models.