IEEE VIS 2025 Supplemental Materials - LVLMs and Aesthetics

This repository contains supplementary materials for our study on Large Vision Language Models (LVLMs) and their effectiveness in evaluating data visualizations. The dataset includes student-created visualizations, expert annotations, model evaluations, and training data used for Retrieval-Augmented Generation (RAG).The repository is organized into the following folders:📁 <b>Data Visualizations</b>Contains <b>11 student-created data visualizations</b> used in the study.Each visualization serves as input for LVLM evaluations.📁 <b>Evaluation Results</b>Includes <b>annotations</b> for each visualization from:Expert evaluation ("ground truth").10 LVLMs (both base and RAG variants).Evaluations assess alignment with visualization principles, interpretability, and coherence.📁 <b>Training Data</b>Contains Tufte &amp; Wilkinson’s books used for Retrieval-Augmented Generation (RAG).These texts provide background knowledge for models incorporating retrieval-based improvements to the base models.