Quantifying the Cultural Recognition Gap in Vision Foundation Models: A Difficulty-Matched Indian Food Benchmark

This repository contains the research code and experimental Jupyter notebooks associated with the paper “Quantifying the Cultural Recognition Gap in Vision Foundation Models: A Difficulty-Matched Indian Food Benchmark.” The study evaluates the ability of vision foundation models to recognize culturally specific Indian food categories using a difficulty-matched benchmark. The repository includes the notebooks used for experimentation, evaluation, and analysis. The experiments are conducted using the Indian Food Image Dataset, which is published separately under its own DOI. The dataset is not included in this repository. This record contains only the research paper-related code and computational materials, while the dataset is maintained as a separate research artifact. The repository is provided to support reproducibility and facilitate further research on cultural representation and recognition in vision foundation models.