PiotrSty's picture
Preserve full print engine with explicit contemporary-document support and evidence
34d78c3 verified
Raw History Blame Contribute Delete
6.29 kB
"""CLI silnika OCR.
Komendy:
ocr recognize <image> [--lang pl|en] [--json] [--correct] ...
ocr check-fabryka [--fabryka-model MODEL]
"""
from __future__ import annotations
import argparse
import json
import sys
from pathlib import Path
from .config import OcrConfig
from .pipeline import OcrEngine
from .postprocess import TextCorrector
def _build_parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(prog="ocr", description="Silnik OCR (CRAFT + TrOCR)")
sub = parser.add_subparsers(dest="command", required=True)
# recognize
rec = sub.add_parser("recognize", help="Rozpoznaj tekst z obrazu lub PDF")
rec.add_argument("image", type=Path, help="艢cie偶ka do obrazu lub PDF")
rec.add_argument("--lang", choices=["pl", "en"], default=None, help="Wymu艣 j臋zyk")
rec.add_argument("--json", action="store_true", help="Wyj艣cie JSON")
rec.add_argument("--no-deskew", action="store_true", help="Pomi艅 korekt臋 pochylenia")
rec.add_argument("--pages", default=None,
help="Zakres stron PDF, np. '1-3,5' (domy艣lnie: wszystkie)")
rec.add_argument("--dpi", type=int, default=300, help="DPI renderowania stron PDF")
rec.add_argument("--correct", action=argparse.BooleanOptionalAction, default=None,
help="Korekta tekstu przez Fabryka API (wymaga FABRYKA_API_KEY)")
rec.add_argument("--correct-low-only", action=argparse.BooleanOptionalAction, default=None,
help="Korekta tylko linii z confidence < --confidence-threshold")
rec.add_argument("--confidence-threshold", type=float, default=None,
help="Pr贸g niskiego confidence (flaga w JSON / selektywna korekta)")
rec.add_argument("--fabryka-model", default=None,
help="Model Fabryka do korekty (domy艣lnie: bielik-11b-v3)")
rec.add_argument("--device", choices=["auto", "cpu", "cuda"], default=None)
rec.add_argument("--backend", choices=["trocr", "paddlevl", "kraken", "unlimitedocr", "auto"], default=None,
help="Backend: trocr, paddlevl, kraken, unlimitedocr (pe艂na strona, CUDA), auto")
rec.add_argument("--max-new-tokens", type=int, default=None,
help="Limit wygenerowanych token贸w Unlimited-OCR (domy艣lnie 4096)")
# Jev (TypeSafe AI)
rec.add_argument("--jev-route", action=argparse.BooleanOptionalAction, default=None,
help="Jev routing backendu (wymaga TYPESAFE_API_KEY)")
rec.add_argument("--jev-score", action=argparse.BooleanOptionalAction, default=None,
help="Jev scoring jako艣ci linii OCR")
rec.add_argument("--jev-validate", action=argparse.BooleanOptionalAction, default=None,
help="Jev walidacja korekty Fabryka")
rec.add_argument("--jev-quality-threshold", type=float, default=None,
help="Pr贸g jako艣ci Jev (0-2) do flagowania linii")
# check-fabryka
chk = sub.add_parser("check-fabryka", help="Sprawd藕 po艂膮czenie z Fabryka API")
chk.add_argument("--fabryka-model", default=None,
help="Model do przetestowania (domy艣lnie: bielik-11b-v3)")
return parser
def _cmd_recognize(args) -> int:
if not args.image.exists():
print(f"Plik nie istnieje: {args.image}", file=sys.stderr)
return 2
config = OcrConfig.from_env()
for arg, field in (("lang", "force_language"), ("correct", "correct_text"),
("correct_low_only", "correct_low_confidence_only"),
("confidence_threshold", "confidence_threshold"),
("fabryka_model", "fabryka_model"), ("device", "device"),
("backend", "recognizer_backend"),
("max_new_tokens", "unlimited_max_new_tokens"),
("jev_route", "jev_route_backend"),
("jev_score", "jev_score_lines"),
("jev_validate", "jev_validate_correction"),
("jev_quality_threshold", "jev_quality_threshold")):
value = getattr(args, arg)
if value is not None:
setattr(config, field, value)
if args.no_deskew:
config.deskew = False
if args.correct_low_only:
config.correct_text = True
if config.correct_low_confidence_only and config.confidence_threshold <= 0:
raise ValueError("Selective correction requires confidence-threshold > 0")
with OcrEngine(config) as engine:
if args.image.suffix.lower() == ".pdf":
results = engine.recognize_pdf(args.image, pages=args.pages, dpi=args.dpi)
else:
results = [engine.recognize(args.image)]
if args.json:
out = [r.to_dict(config.confidence_threshold) for r in results]
print(json.dumps(out[0] if len(out) == 1 else out, ensure_ascii=False, indent=2))
else:
for i, r in enumerate(results):
if len(results) > 1:
print(f"--- strona {r.page_number or i + 1} ---")
print(r.text)
return 0
def _cmd_check_fabryka(args) -> int:
config = OcrConfig.from_env()
if args.fabryka_model is not None:
config.fabryka_model = args.fabryka_model
corrector = TextCorrector(config)
print(f"Endpoint: {corrector.base_url}")
print(f"Model: {config.fabryka_model}")
key = corrector.api_key
if key:
# poka偶 tylko prefix ze wzgl臋d贸w bezpiecze艅stwa
print(f"Klucz: {key[:12]}...{key[-4:]} (d艂ugo艣膰: {len(key)})")
else:
print("Klucz: BRAK (ustaw FABRYKA_API_KEY)")
return 1
print()
ok, msg = corrector.check_connection()
if ok:
print(f"[OK] {msg}")
return 0
else:
print(f"[FAIL] {msg}")
return 1
def main(argv: list[str] | None = None) -> int:
parser = _build_parser()
args = parser.parse_args(argv)
if args.command == "recognize":
return _cmd_recognize(args)
if args.command == "check-fabryka":
return _cmd_check_fabryka(args)
return 1
if __name__ == "__main__":
raise SystemExit(main())