Pipeline de datasets (ingesta → curado → export → push a Hugging Face).
make setup # instala dependencias (venv + libs)
make ingest # copia imágenes desde fuentes locales/Drive
make curate # dedup + normaliza + meta.csv
make export # crea train/val/test en data_splits
make hfpush # sube a Hugging Face (ajusta HF_REPO en Makefile)
1sed -n '1,80p' README.md
2
