A Semi-Automatic LLM-Assisted Process for Converting 2D Graph Images into 3D-Printable Tactile Models for Blind and Low-Vision Learners

This repository accompanies research presenting a semi-automatic pipeline for converting 2D graph images into 3D-printable tactile models. The workflow uses Large Language Models (LLMs) to support image-to-3D translation and enable iterative refinement when errors occur. The goal is to help educators and practitioners without 3D modeling expertise produce tactile-accessible graphs for Blind and Low-Vision (BLV) learners. Contents of this archive include: 2D graph image dataset, organized in subfolders (see "Dataset organization" below). Process prompts (TXT) used to execute the different phases of the process. Correction prompts (TXT) designed to be applied when spiecific issues arise during execution, supporting iterative error fixing. Python script (.py) used to randomly generate the 2D graph images included in the dataset. User guide (PDF) providing step-by-step instructions to run the process, targeted at non-expert users. Dataset organization. The dataset folder is structured into two main subfolders: directed graphs and undirected graphs. For each graph type, the dataset includes two graph sizes (5 nodes and 10 nodes) with increasing complexity. Specifically, 5-node graphs include instances with 6, 7, and 8 edges, while 10-node graphs include instances with 11, 12, and 13 edges. This controlled design provides systematic variations in size and density to support evaluation and reuse across multiple levels of graph complexity. The materials are released to support reproducibility and to facilitate reuse and extension of the proposed process in accessibility-oriented visualization and 3D printing contexts.