Computational and Data-driven Design Spaces: From designing shells and elastic curves to appearance editing of human faces in GAN space
Abstract
In this talk, I will first describe recent progress in engineering design toward novel concepts for modeling, designing, and reproducing objects with nontrivial shapes, topologies, and functionalities. I will start by highlighting how data-driven techniques can enable the interactive design of cold-bent glass façades that can be seamlessly integrated into a typical architectural design pipeline. Making a step towards robotic materials, I will then introduce novel approaches for discovering and designing architected materials and demonstrate their applicability for encoding temporal shape evolution in architected shells that assume complex shapes and doubly curved geometries. Switching gears, I will also touch on the appearance editing of head portraits. I will demonstrate an approach that operates in the generative model space and learns transformations in the latent space of StyleGAN. This combines the best of supervised learning and generative adversarial modeling, produces high-quality photorealistic results for in-the-wild images, can edit the illumination and pose simultaneously, and runs at interactive rates. Finally, I will reflect on the successes and challenges of data-driven design, contrast this approach with our most recent work on the rigorous geometric characterization of the planar elastic rods' design space, and discuss opportunities for further work in this area.
