3D Shape Generation

3D shape generation studies how to create complete, useful 3D objects and scenes, supporting tasks from design and animation to medical modeling and robotics. Common approaches include diffusion and flow-matching models, autoregressive token generators, and neural representations of point clouds, voxels, meshes, implicit surfaces, or CAD geometry. Models may be conditioned on text, images, multiple views, partial shapes, or part layouts, and are evaluated on benchmarks such as ShapeNet, ModelNet, and Objaverse for fidelity, diversity, and alignment. A central challenge is balancing visual quality and fast generation with geometric validity, coherent parts, controllability, and reliable completion from sparse or conflicting inputs.