VBIG Benchmark Dataset

This dataset accompanies the master's thesis Automated BI Dashboard Narration: An Empirical Study of Small Vision-Language Models for Multi-Level Semantic Insight Generation. The dataset contains four files: dashboard metadata, human-written ground-truth annotations, VLM-generated outputs, and evaluation scores produced from evaluating seven open-source small Vision-Language Models (≤3B parameters) on 40 BI dashboard images sourced from Tableau Public using the Superstore dataset. A README.md file is included that describes the structure and column definitions for each file. Full methodological details are available in the thesis.