MeshCoder is a multimodal dataset for generating editable code from 3D point clouds, released in 2025 by the Shanghai Artificial Intelligence Laboratory in collaboration with Tsinghua University, Harbin Institute of Technology (Shenzhen), and other institutions. The related research paper is titled "...".MeshCoder: LLM-Powered Structured Mesh Code Generation from Point CloudsThe goal is to promote the development of large language models in 3D scene parsing, structural understanding, and programmable geometric reconstruction.

This dataset contains 100,000 object-code pairing samples, covering more than 40 classes of artificial objects with complex structures. Each object contains more than 100 semantic components, and provides corresponding editable Blender Python scripts for the model, supporting code-based shape and topology modifications.

Its core structural elements include:

• 3D point cloud data: a three-dimensional geometric representation of a real or synthetic object.
• Semantic component annotation: Structural information divided by components
• The corresponding Blender Python program can directly reconstruct the shape and perform topological transformations.