High-Resolution granular modeling for style transfer and 3D brushstrokes

Keywords
Loading...
Thumbnail Image
Issue Date
2021-06-24
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
We investigate the e ects of granular factorization over style transfer and depth estimation. Previous work on style transfer focuses mainly on creating high-quality images; this is only part of the picture. We consider that a proper style transfer model would require 3D brush strokes and textures. However, the most similar work for 3D brushstrokes focuses on assisting the artist and not on a fully autonomous end-to-end model. Our approach requires a granular factorization to control the scale of the style and brushstrokes. This factorization also allows our models to create ultrahigh- de nition images. We present a framework of two sequential models that extends the idea of style transfer by including a painter's 3D brushstrokes and textures. We show that our models can create a 3D depth map out of 2D brushstrokes. With the use of granular factorization our model is capable of controlling the scale of the style transfer and the preservation of original features in the content image.
Description
Citation
Faculty
Faculteit der Sociale Wetenschappen