ShapeLib: Using LLMs to Design Programmatic 3D Shape Abstraction Libraries
By
[Submitted on 13 Feb 2025 (v1), last revised 31 May 2026 (this version, v3)]
Properly proved. Has structure, has flavour, has a point.
Summary
ShapeLib is a novel method that leverages Large Language Models (LLMs) to design libraries of programmatic 3D shape abstractions. The system accepts user-provided design intent through text descriptions and exemplar shapes, then uses a guided LLM workflow to propose, implement, and validate abstraction functions. It develops library-specific recognition networks to map shapes to programs using these abstractions, enabling generalization beyond seed sets. The method combines LLM reasoning with geometric reasoning to author reusable, interpretable shape abstractions that support downstream applications like shape editing and generation.
Key quotes
· 5 pulledWe present ShapeLib, the first method that uses the priors of Large Language Models (LLMs) to design libraries of programmatic 3D shape abstractions.
Our framework takes a step towards realizing the long-standing shape analysis aspiration of discovering reusable, programmatic shape abstractions while exposing interpretable, semantically aligned interfaces.
Across multiple modeling domains (split by shape category), we find that LLMs, when thoughtfully combined with geometric reasoning, can be guided to author libraries of abstraction functions that generalize across shape distributions.
Our extensive evaluation demonstrates that ShapeLib provides distinct advantages over prior alternative abstraction discovery works in terms of generalization, usability, and maintaining plausibility under manipulation.
ShapeLib's abstraction functions unlock a number of downstream applications, combining LLM reasoning over shape programs with geometry processing tools to support shape editing and generation workflows.
You might also wanna read
LL3M: Large Language Models for 3D Asset Creation in Blender
LL3M (Large Language 3D Modelers) is a system developed by the University of Chicago that employs large language models to generate Python c
ShapeR: Robust Conditional 3D Shape Generation from Casual Image Captures
ShapeR is a novel system for generative, object-centric 3D reconstruction from casual image sequences. The research paper presents a method
Mesh-LLM: Distributed LLM Inference System Using llama.cpp Across Multiple Machines
Mesh-LLM is a reference implementation that enables distributed inference of large language models across multiple machines by compiling lla
github.com·2mo ago
Enhancing Abstraction in Large Language Models Through Nature-Inspired Semantic Patterns
ieeexplore.ieee.org·11mo ago
"Large Model Data Engineering: Architecture, Algorithms and Practical Projects" - A Comprehensive Guide to LLM Data Engineering
This is a data engineering book focused on large language model (LLM) data engineering, covering architecture, algorithms, and practical pro
github.com·3mo ago
High-Fidelity 3D Shape Generation Using Scalable Geometric Refinement
The article discusses a high-fidelity 3D shape generation pipeline that uses scalable geometric refinement techniques. It appears to focus o
pku-yuangroup.github.io·5mo ago