Modeling, Capturing, and Understanding Animals in 3D

Silvia Zuffi

IMATI-CNR, IT

Abstract

Over the past decade, computer vision has made remarkable progress in reconstructing humans in 3D from images and videos. Extending these advances to animals opens exciting opportunities but also introduces new challenges due to the large diversity of species, body shapes, articulated structures, appearances, and the limited availability of annotated data. In this lecture, I will review recent progress in modeling and reconstructing animals in 3D, highlighting how ideas originally developed for human body modeling can be adapted to animals. I will introduce parametric models of animal shape and pose and discuss methods for estimating 3D animal geometry, pose and motion from monocular images and videos. Beyond reconstruction, these models provide a structured representation that enables a deeper understanding of animal morphology, pose, and movement across species. The lecture will also cover the datasets and supervision strategies used to train such systems, as well as techniques for transferring knowledge across species and handling challenges such as fur, appearance variation, and occlusions. Finally, I will discuss open problems and emerging research directions toward more general and scalable approaches for capturing and understanding animals in the wild.