PhotoHandler: Manipulation of Portrait Images with StyleGANs Using Text

Castillo-Arredond G.; Moreno-Carhuacusma D.; Ugarte W. · ICSBT International Conference on Smart Business Technologies (2023) · DOI: 10.5220/0012123200003552

Resumen

There are many methods and approaches used for manipulation of images. However, according to the state of the art, the ones related to StyleGANs has shown better overall results when it comes to this task. In this work, we proposed to use StyleGANs in order to allow the manipulation of portrait images using text. We used a method called StyleCLIP which combine StyleGANs with CLIP, a neural network that connects text and images. This method has two numeric parameters, which are the manipulation strength and the disentanglement threshold. The contribution consists of adjusting these parameters based on the entered text in order to improve the result of the edited image. With this technology, we built an app that will help people with poor digital skills to edit their photos easily by using words. The opinion of the users of the application has been validated through a survey. The results obtained allowed us to demonstrate that our method make a satisfactory image edition for most users.

Instituciones: Universidad Peruana de Ciencias Aplicadas