Title: Parametric-Based Models for Artistic Representations
Name: Manuel Rodríguez Ladrón de Guevara, PhD Candidate in Computational Design
Date: Friday, January 19, 2024
Time: 11:00am-1:00pm ET
Location: Margaret Morrison Carnegie Hall (MMCH), Room A11 and on Zoom
Thesis Committee:
Ramesh Krishnamurti (Chair)
Emeritus Professor
School of Architecture, Carnegie Mellon University
Daragh Byrne
Associate Teaching Professor
School of Architecture, Carnegie Mellon University
Jun-Yan Zhu
Robotics Institute
Carnegie Mellon University
Abstract:
Driven by the popular adoption of AI for artistic purposes, this research examines its technical and ethical implications, and presents new approaches for parametric algorithms that generate different type of artistic representations, based on optimization methods and learning models. In the midst of the success of machine learning algorithms in image generation and other creative tasks of pixel-based diffusion and large language models, my dissertation is framed within the field of parametric representations, which is closer to human art than pixel-based methods. An array of computational methods including procedural, optimization, and machine learning are analyzed and proposed with the intention of disseminating the generation of artwork by computers.
To understand and situate the current AI models used for art, we provide a comprehensive review and implementation of artistic computational methods, ranging from classical procedural hand-made, rule-based algorithms, to the most advanced AI methods. After analyzing the technical possibilities of existing methods, we present new algorithms that focus on particular gaps in the literature.
Particularly, this research studies three main problems that exist in the literature: stylization, controllability, and identity preservation. Stylization, the process of applying a particular artistic style to an image, is a common subject within generative algorithms, and key to artistic success. Controllability enables intentional painting processes and thus, it is a steppingstone for further stylization. Identity preservation is the ability for a learning model to preserve key content features from the original image through the artistic representation process. Finding the right combination of stroke primitives for a particular artistic style and, by extension, for a painting strategy that leads to certain styles is not fully resolved under a parametric framework, as there exists a trade-off between reconstructions of the input image and a controllable stylistic variation. Existing works that address style are still limited in style variations and controllability, and principally use different stroke models and textures to output styles. State-of-the-art algorithms normally output strokes in an uncontrollable manner without a planned strategy that might help stylization. However, human artists employ painting techniques such as “blocking in”, grouping by semantics or colors, “background-foreground” or “color-then- contours” that help them convey artistic styles.
Throughout the different algorithms presented, we demonstrate new ways to find stylization, controllability, and identity preservation. We disentangle such a complex landscape of artistic styles and strategies, and leverage some artistic vision under some perception of art. We finally tap into computational creativity, whether algorithms can be creative, and discuss future steps in the field of machine learning and art.