Volumetric Primitive

Volumetric primitives represent 3D shapes using simple geometric building blocks like cuboids, cylinders, or superquadrics, aiming for efficient and compact scene representations. Current research focuses on developing robust algorithms for fitting these primitives to complex data, including neural network-guided approaches and diffusion models operating directly on primitive parameters, improving accuracy and speed compared to traditional methods. This approach offers significant advantages in applications ranging from 3D modeling and rendering to real-time simulations and scene understanding, enabling faster processing and more efficient data storage for complex 3D environments.

Papers