ShapeLib: Designing a library of programmatic 3D shape abstractions with Large Language Models
Researchers have developed ShapeLib, a novel system that leverages Large Language Models (LLMs) to create libraries of programmatic 3D shape abstractions. This method accepts user intent through text descriptions and exemplar shapes, guiding the LLM to propose and validate functions for the library. The system then employs library-specific recognition networks to map various shape representations to programs utilizing these new abstractions, demonstrating a step towards reusable and semantically aligned shape analysis interfaces. AI
IMPACT Enables creation of reusable programmatic shape abstractions, unlocking downstream applications in shape editing and generation.