Physically-based Generative Neural Graphics

Compared to the extensive time and computational resources once required for intricate details like Merida's hairstyle or the ocean animations in Pirates of the Caribbean, emerging AI-powered graphics algorithms are revolutionizing the field, making advanced visual effects more accessible to everyday content creators than ever before. This VIP project aims to invent novel algorithms by bringing together the best of two worlds: traditional, physics-based simulation methods and the fast-evolving field of AI-driven dynamic graphics. The mission of this project stands at a unique intersection of multiple disciplines, from computer graphics and computational physics to machine learning and data science, presenting a unique opportunity to explore the best practice of integrating physical authenticity into AI-generated graphics. By developing novel data structures, PDE solvers, neural representations, and machine learning paradigms, we aim to significantly elevate the state of the art in physically-based animation and pave the way for innovations that could revolutionize how we create and interact with the physical world.