# KBL / Royal Danish Library open-image harvesting

This folder contains scripts for harvesting open-license image material from the Royal Danish Library
COP/Digital Collections API for Danish CLIP training.

Included by default:

- `Billeder`
- `Kort og Atlas`, kept separable with `source_subset=maps`

Excluded before scraping:

- aerial photographs / `Danmark set fra luften`
- `Danske aviser 1666-1883`
- text-heavy COP editions: books, letters, manuscripts, pamphlets, printed matter and catalogues

The scraper over-collects records using license keywords, then filters by actual MODS
`accessCondition` rights text. KBL commonly expresses public-domain/no-known-rights material as
`Materialet er fri af ophavsret`.

## Commands

```powershell
cd "c:\Users\V4lde\Downloads\DanishCLIP\new datasets\kbl"

python scrape_kbl_cop_metadata.py --out-dir data/kbl_open_metadata

$env:HF_TOKEN = "<set outside source control>"
python create_dataset_card.py --repo-id "<user>/kbl-open-images" --metadata-summary data/kbl_open_metadata/summary.json --out data/README.md --upload
python upload_kbl_metadata.py --repo-id "<user>/kbl-open-images" --metadata-dir data/kbl_open_metadata

python build_upload_kbl_images.py --repo-id "<user>/kbl-open-images" --metadata-jsonl data/kbl_open_metadata/kbl_open_metadata.jsonl --work-dir data/kbl_image_work --upload --skip-existing --records-per-shard 500 --download-workers 12
```

Do not commit or paste the Hugging Face token into any file.