Speaker: Aysegul Dundar

Title: 3D-Aware Generative Modeling for Head Reconstruction, Editing, and Stylization

Abstract: Generative models have reached remarkable visual fidelity in 2D image synthesis; however, extending this realism to 3D-consistent head reconstruction, editing, and stylization remains fundamentally challenging. In particular, identity preservation, multiview consistency, and geometric coherence introduce constraints that standard diffusion and GAN-based approaches are not explicitly designed to address.

In this talk, I will present a set of methods for 3D-aware generative modeling that combine structured 3D representations with modern generative priors. By integrating these representations with diffusion and inversion techniques, we enable high-fidelity head reconstruction from single images, reference-guided 3D editing, and identity-preserving stylization that remains consistent across viewpoints, providing a practical pathway toward controllable and geometry-aware 3D synthesis.

We look forward to seeing you there!

https://us06web.zoom.us/j/84636366540

MEETING ID: 84636366540
PASSCODE: 25862