# Data

Download from Kaggle (needs an API token from https://www.kaggle.com/settings):

    pip install kaggle
    python data/download.py

Expected layout after download:

    data/artwork_metadata.csv   3,997 rows — filename, religion (1,000 each of
                                buddhism / christianity / hinduism; 997 islam),
                                sub_religion, artist, title, year, place,
                                source, source_id, source_url, image_url
    data/images/                the 3,997 images, filenames match the CSV
                                (download.py flattens Kaggle's images/images/)

Generated later by the pipeline (gitignored):

    data/masks/         guarded background masks + verdicts.csv
    data/features/      one parquet per feature family