polocrbench-train-pool-v1 / integrate_gardnotes.py
PiotrSty's picture
Upload folder using huggingface_hub
359ceb1 verified
Raw History Blame Contribute Delete
4.31 kB
"""Integrate the self-licensed gardening notes (1988-89 garden journal) into
polocrbench-train-pool-v1 (handwriting-gardnotes source).
License holder: Piotr Styła (declared CC BY 4.0, 2026-10-23).
Content: handwritten garden journal on yellowed paper (weather table, plant
and variety lists with pencil/pen strike-outs, transplanting and fertilizing
log). Some pages cut at edges ([ucięte]); struck text excluded per benchmark
convention ([skreślone] marker). One re-scan pair (Scan_0012 / Scan_0017 =
same source page, different crops) is kept deliberately as rescanning
variety. No personal identifiers found (deterministic PII scan).
"""
import collections
import hashlib
import json
import shutil
from pathlib import Path
BASE = Path('C:/Users/Hipek/OCR_external_data')
WORK = BASE / 'skany-antoni-work'
POOL = BASE / 'polocrbench-train-pool-v1'
RESCAN_PAIR = {'Scan_0012 (1)', 'Scan_0017 (1)'}
def sha(data):
return hashlib.sha256(data).hexdigest()
def main():
trans = {}
for line in (WORK / 'transcriptions.jsonl').read_text(encoding='utf-8').splitlines():
row = json.loads(line)
trans[row['page_id']] = row['text']
originals = {p.stem: sha(p.read_bytes()) for p in (WORK / 'originals').glob('*.jpg')}
assert len(trans) == 10
rows = []
for stem in sorted(trans):
src = WORK / 'pages' / f'{stem}.jpg'
pool_id = 'hwgard_' + stem.lower().replace(' ', '_').replace('(', '').replace(')', '')
data = src.read_bytes()
shutil.copyfile(src, POOL / 'images' / f'{pool_id}.jpg')
rows.append({'id': pool_id, 'source': 'handwriting-gardnotes',
'source_ref': 'self-licensed garden journal 1988-89 '
'(license holder: Piotr Styła, CC BY 4.0, 2026-10-23)',
'image': f'images/{pool_id}.jpg', 'image_sha256': sha(data),
'text': trans[stem], 'text_sha256': sha(trans[stem].encode()),
'width': 700, 'height': 1000, 'split': 'train',
'license': 'cc-by-4.0', 'human_verified': False, 'gold': False,
'page_fragment': True,
'rescan_group': 'rescan-pair-1' if stem in RESCAN_PAIR else None,
'original_image_sha256': originals.get(stem),
'transcription': 'ai-preannotation-cut-markers'})
manifest_path = POOL / 'manifest.jsonl'
existing = manifest_path.read_text(encoding='utf-8')
manifest_path.write_text(existing + ''.join(
json.dumps(r, ensure_ascii=False) + '\n' for r in rows), encoding='utf-8')
report = json.loads((POOL / 'report.json').read_text(encoding='utf-8'))
all_rows = [json.loads(s) for s in manifest_path.read_text(encoding='utf-8').splitlines()]
report['schema'] = 'polocrbench-train-pool-v1.5'
report['pages'] = len(all_rows)
report['by_source'] = dict(collections.Counter(r['source'] for r in all_rows))
report['by_split'] = dict(collections.Counter(r['split'] for r in all_rows))
report['handwriting_gardnotes_pages'] = len(rows)
report['handwriting_gardnotes_license'] = 'CC BY 4.0 (license holder: Piotr Styła)'
report['handwriting_gardnotes_content'] = ('garden journal 1988-89 on yellowed paper; '
'plant/variety lists with strike-outs, weather '
'table, gardening log; one re-scan pair kept '
'as rescanning variety (rescan_group field)')
report['licenses']['handwriting-gardnotes'] = 'CC BY 4.0 (license holder: Piotr Styła)'
(POOL / 'report.json').write_text(json.dumps(report, ensure_ascii=False, indent=2),
encoding='utf-8')
sums = {p.relative_to(POOL).as_posix(): sha(p.read_bytes())
for p in sorted(POOL.rglob('*')) if p.is_file() and p.name != 'checksums.json'}
(POOL / 'checksums.json').write_text(json.dumps(sums, indent=2), encoding='utf-8')
print(json.dumps({'gardnotes_rows': len(rows), 'pool_pages': len(all_rows),
'by_source': report['by_source'], 'checksum_entries': len(sums)},
ensure_ascii=False, indent=2))
if __name__ == '__main__':
main()