3D Primitive

3D primitive-based modeling focuses on representing complex 3D shapes as assemblies of simpler geometric primitives, such as superquadrics or convex quadrics, for efficient representation and manipulation. Current research emphasizes learning-based approaches, often employing neural networks (e.g., differentiable rendering frameworks and neural architecture search) to optimize primitive parameters from 2D or sparse 3D data, enabling tasks like 3D reconstruction, scene understanding, and shape segmentation. This approach offers advantages in data efficiency, computational speed, and interpretability compared to high-fidelity methods, impacting fields like computer graphics, robotics, and 3D scene understanding through improved efficiency and the creation of more easily manipulated 3D models.