Neural Clothing on Mobile
Arm
Abstract
Realistic dynamic garment deformation on arbitrary meshes remains challenging, as existing Machine Learning approaches often compromise either on quality of the deformation, generalization or processing requirement. This makes dynamic garment deformation unsuitable for real-time applications on portable and handheld/mobile devices.
In this talk we will discuss some of the challenges when creating compelling clothing behaviour for mobile and ways to address them. We will present an optimized approach using a Graph Attention Network able to predict static and dynamic cloth deformations of worn garments. Our method addresses the performance limitations of Graph Neural Networks, while taking advantage of multi-level message passing and unsupervised learning. We review the requirements for a wide adoption of neural clothing models in mobile devices and possible future directions.
Mina Dimova
Bio
Mina Dimova is a Principal Software Engineer at Arm where she has worked in multiple teams on Research and Development in graphics and machine learning.
She is currently working in Developer Advocacy exploring emerging use-cases for showcasing Arm’s latest technology and helping shape future hardware generations.
She has a MSc in Computer Graphics, Vision and Imaging from UCL, London.
Contact
Send questions to general@i3dsymposium.org for general inquiries, registration, and sponsorship.
Direct queries about paper submissions to papers@i3dsymposium.org and poster submissions to posters@i3dsymposium.org.