Keynote Talk 1: Dr. Silvia Zuffi

Cross-Species Statistical Models for 3D Animal Model-Based Reconstruction

Reconstructing 3D animal shape and pose from images is challenging due to limited 3D data and large morphological variation across species. I will present our work on multi-species statistical shape models that enable controllable and realistic 3D animal representations. This includes SMAL (Skinned Multi-Animal Linear Model), the first parametric model capturing shape variation across quadrupeds, and its use in estimating animal geometry directly from visual data. I will also introduce AWOL (Analysis WithOut synthesis using Language), a framework that leverages natural language to guide parametric 3D models, improving controllability and interpretability. Finally, I will show how AWOL-generated synthetic data can support training systems that predict 3D shape and pose for diverse animal species.