Image-to-Text
Transformers
Safetensors
Turkish
English
ocr
turkish
document-intelligence
vision-language-model
svtrv2
qwen2.5-vl
kvkk
pii-redaction
legal-nlp
ictihat
financial-verification
visual-redaction
Eval Results (legacy)
Instructions to use ersanbil/TurkceOCR with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use ersanbil/TurkceOCR with Transformers:
# Use a pipeline as a high-level helper # Warning: Pipeline type "image-to-text" is no longer supported in transformers v5. # You must load the model directly (see below) or downgrade to v4.x with: # pip install "transformers<5.0.0" from transformers import pipeline pipe = pipeline("image-to-text", model="ersanbil/TurkceOCR")# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("ersanbil/TurkceOCR", device_map="auto") - Notebooks
- Google Colab
- Kaggle
File size: 13,970 Bytes
c8b6043 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 | """
TürkçeOCR (`turkceocr.server` / `turkceocr.server`) FastAPI REST Çıkarım ve Doğrulama Servisi.
Uç Noktalar (Endpoints):
- `GET /health` : Model ailesi durumu, özel HF depoları ve 13 yapısal özel token.
- `POST /v1/verify` : Ham VLM/OCR Markdown metnini `TurkceVerify` (TCKN/VKN/IBAN/İ-ı) ve
`TurkishFinancialSolver` (Matrah + KDV = Genel Toplam) ile onarır.
- `POST /v1/ocr/demo` : Gerçekçi kaşeli/bozulmuş Türkçe belge sentezler, `HighResDiacriticTiler`
ile yamalara böler, `TurkceOCRLine` ve `TurkceVerify` hattından geçirir.
- `POST /v1/ocr/image` : Base64 kodlu görüntüyü yüksek çözünürlüklü yamalara böler,
`TurkceOCRLine` satır okumasını ve `TurkceVerify` onarımını çalıştırır.
"""
from __future__ import annotations
import argparse
import base64
import io
from typing import Any, Dict, List, Optional
from fastapi import FastAPI, HTTPException
from PIL import Image
from pydantic import BaseModel, Field
from turkceocr.cli import inject_realistic_ocr_corruptions, synthesize_demo_document
from turkceocr.models.vl_model import SUPPORTED_VL_BACKBONES, TURKCEOCR_SPECIAL_TOKENS
from turkceocr.pipeline import TurkceOCRPipeline
from turkceocr.postprocess.entity_engine import UnifiedTurkishEntityEngine
from turkceocr.postprocess.visual_redactor import VisualRedactor
TURKCEOCR_HF_REPOS: List[str] = [
"ersanbil/TurkceOCR",
"ersanbil/kvkk-pii-ner",
"ersanbil/TurkceOCR-Bench",
"ersanbil/TurkceOCR-Demo",
]
class HealthResponse(BaseModel):
status: str = "ok"
brand: str = "TürkçeOCR"
version: str = "0.2.0"
hf_repos: List[str] = Field(default_factory=lambda: list(TURKCEOCR_HF_REPOS))
special_tokens: List[str] = Field(default_factory=lambda: list(TURKCEOCR_SPECIAL_TOKENS))
class VerifyRequest(BaseModel):
markdown: str = Field(..., description="Onarılacak ham OCR / VLM Markdown metni.")
repair_financials: bool = Field(
default=True,
description="Matrah + KDV = Genel Toplam cebirsel ve görsel denklem çözücüsünü çalıştır.",
)
backbone: str = Field(default="lightonocr", description="Kullanılacak VLM yapılandırması.")
class DemoOCRRequest(BaseModel):
doc_type: str = Field(
default="erp_table",
description="Sentezlenecek belge tipi: erp_table, invoice, pos, handwritten, gazette, born_digital_2col.",
)
apply_degradation: bool = Field(
default=True,
description="TurkceDegrade kaşe, imza ve fiziksel bozulma motorunu uygula.",
)
backbone: str = Field(default="lightonocr", description="lightonocr veya paddleocr_vl.")
seed: int = Field(default=2026, description="Rastgelelik tohumu.")
class ImageOCRRequest(BaseModel):
image_base64: str = Field(..., description="Base64 kodlu PNG/JPEG belge görüntüsü.")
raw_markdown: Optional[str] = Field(
default=None,
description="İsteğe bağlı ön-VLM Markdown çıktısı (verilmezse satır kırpımlarından üretilir).",
)
backbone: str = Field(default="lightonocr", description="lightonocr veya paddleocr_vl.")
repair_financials: bool = Field(default=True, description="Finansal denklem onarımını çalıştır.")
class RedactDocumentRequest(BaseModel):
markdown: Optional[str] = Field(
default=None,
description="Analiz edilecek ve KVKK kapsamında karartılacak metin/Markdown.",
)
image_base64: Optional[str] = Field(
default=None,
description="İsteğe bağlı Base64 kodlu belge görüntüsü (verilirse 2B #000000 karartma uygulanır).",
)
doc_type: str = Field(
default="invoice",
description="Görüntü veya metin verilmezse sentezlenecek demo belge tipi.",
)
backbone: str = Field(default="lightonocr", description="lightonocr veya paddleocr_vl.")
def _serialize_financial_check(fc: Any) -> Optional[Dict[str, Any]]:
if fc is None:
return None
return {
"is_consistent": bool(fc.is_consistent),
"was_repaired": bool(fc.was_repaired),
"subtotal_matrah": float(fc.subtotal_matrah),
"vat_kdv": float(fc.vat_kdv),
"grand_total": float(fc.grand_total),
"inferred_vat_rate": fc.inferred_vat_rate,
"repair_note": str(fc.repair_note),
}
def create_turkceocr_app() -> FastAPI:
"""TürkçeOCR FastAPI uygulamasını oluşturur ve yapılandırır."""
api = FastAPI(
title="TürkçeOCR Inference & Verification API",
description=(
"İki Katmanlı Açık Kaynak Türkçe OCR ve Doküman Zekası Ailesi "
"(TürkçeOCR-VL-1B, TürkçeOCR-Line SVTRv2-TR ve TurkceVerify)"
),
version="0.2.0",
)
pipeline_cache: Dict[str, TurkceOCRPipeline] = {}
def _get_pipeline(backbone: str = "lightonocr") -> TurkceOCRPipeline:
norm = backbone if backbone in SUPPORTED_VL_BACKBONES else "lightonocr"
if norm not in pipeline_cache:
pipeline_cache[norm] = TurkceOCRPipeline(vl_backbone=norm)
return pipeline_cache[norm]
@api.get("/health", response_model=HealthResponse)
def health_check() -> Dict[str, Any]:
return {
"status": "ok",
"brand": "TürkçeOCR",
"version": "0.2.0",
"hf_repos": list(TURKCEOCR_HF_REPOS),
"special_tokens": list(TURKCEOCR_SPECIAL_TOKENS),
}
@api.post("/v1/verify")
def verify_markdown(req: VerifyRequest) -> Dict[str, Any]:
pipeline = _get_pipeline(req.backbone)
page_out = pipeline.postprocess_document_markdown(
raw_markdown=req.markdown,
image=None,
repair_financials=req.repair_financials,
)
rep = page_out.verification_report
fin_dict = _serialize_financial_check(page_out.financial_check)
return {
"brand": "TürkçeOCR",
"raw_markdown": page_out.raw_markdown,
"markdown": page_out.markdown,
"verified_markdown": page_out.markdown,
"tckns": rep.valid_tckns,
"valid_tckns": rep.valid_tckns,
"vkns": rep.valid_vkns,
"valid_vkns": rep.valid_vkns,
"ibans": rep.valid_ibans,
"valid_ibans": rep.valid_ibans,
"repaired_fields": rep.repaired_fields,
"financial_check": fin_dict,
}
@api.post("/v1/ocr/demo")
def ocr_demo_document(req: DemoOCRRequest) -> Dict[str, Any]:
try:
sample = synthesize_demo_document(
doc_type=req.doc_type,
apply_degradation=req.apply_degradation,
seed=req.seed,
)
except ValueError as exc:
raise HTTPException(status_code=400, detail=str(exc)) from exc
pipeline = _get_pipeline(req.backbone)
corrupted_md = inject_realistic_ocr_corruptions(sample)
page_out = pipeline.postprocess_document_markdown(
raw_markdown=corrupted_md,
image=sample.image,
repair_financials=True,
)
line_results: List[Dict[str, Any]] = []
for idx, (crop_img, gt_line) in enumerate(sample.line_crops[:4]):
pred_text, conf = pipeline.recognize_line_image(crop_img)
line_results.append(
{
"line_index": idx,
"ground_truth_line": gt_line,
"predicted_text": pred_text,
"confidence": round(float(conf), 4),
}
)
rep = page_out.verification_report
return {
"brand": "TürkçeOCR",
"doc_type": sample.doc_type,
"layout_variant": sample.layout_variant,
"image_size": list(sample.image.size),
"tile_count": len(page_out.tile_bboxes),
"tile_bboxes": [list(b) for b in page_out.tile_bboxes],
"raw_markdown": page_out.raw_markdown,
"markdown": page_out.markdown,
"verified_markdown": page_out.markdown,
"expected_kie": sample.expected_kie,
"tckns": rep.valid_tckns,
"valid_tckns": rep.valid_tckns,
"vkns": rep.valid_vkns,
"valid_vkns": rep.valid_vkns,
"ibans": rep.valid_ibans,
"valid_ibans": rep.valid_ibans,
"repaired_fields": rep.repaired_fields,
"financial_check": _serialize_financial_check(page_out.financial_check),
"line_recognitions": line_results,
}
@api.post("/v1/ocr/image")
def ocr_base64_image(req: ImageOCRRequest) -> Dict[str, Any]:
try:
payload = req.image_base64.split(",")[-1].strip()
raw_bytes = base64.b64decode(payload)
img = Image.open(io.BytesIO(raw_bytes)).convert("RGB")
except Exception as exc:
raise HTTPException(
status_code=400,
detail=f"Geçersiz base64 görüntü verisi: {exc}",
) from exc
pipeline = _get_pipeline(req.backbone)
w, h = img.size
strip_coords = [
(0, int(h * 0.04), w, int(h * 0.16)),
(0, int(h * 0.20), w, int(h * 0.34)),
(0, int(h * 0.68), w, int(h * 0.84)),
]
line_results: List[Dict[str, Any]] = []
decoded_lines: List[str] = []
for idx, bbox in enumerate(strip_coords):
crop = img.crop(bbox)
pred_text, conf = pipeline.recognize_line_image(crop)
line_results.append(
{
"line_index": idx,
"bbox": list(bbox),
"predicted_text": pred_text,
"confidence": round(float(conf), 4),
}
)
if pred_text.strip():
decoded_lines.append(pred_text.strip())
effective_raw_md = (
req.raw_markdown
if req.raw_markdown is not None
else ("\n".join(decoded_lines) or "# TürkçeOCR Görüntü Çıktısı")
)
page_out = pipeline.postprocess_document_markdown(
raw_markdown=effective_raw_md,
image=img,
repair_financials=req.repair_financials,
)
rep = page_out.verification_report
return {
"brand": "TürkçeOCR",
"image_size": [w, h],
"tile_count": len(page_out.tile_bboxes),
"tile_bboxes": [list(b) for b in page_out.tile_bboxes],
"raw_markdown": page_out.raw_markdown,
"markdown": page_out.markdown,
"verified_markdown": page_out.markdown,
"tckns": rep.valid_tckns,
"valid_tckns": rep.valid_tckns,
"vkns": rep.valid_vkns,
"valid_vkns": rep.valid_vkns,
"ibans": rep.valid_ibans,
"valid_ibans": rep.valid_ibans,
"repaired_fields": rep.repaired_fields,
"financial_check": _serialize_financial_check(page_out.financial_check),
"line_recognitions": line_results,
}
entity_engine = UnifiedTurkishEntityEngine(use_ner=False)
visual_redactor = VisualRedactor(entity_engine=entity_engine)
@api.post("/v1/ocr/redact")
def redact_ocr_document(req: RedactDocumentRequest) -> Dict[str, Any]:
pipeline = _get_pipeline(req.backbone)
img: Optional[Image.Image] = None
raw_md: str = req.markdown or ""
if req.image_base64:
try:
payload = req.image_base64.split(",")[-1].strip()
raw_bytes = base64.b64decode(payload)
img = Image.open(io.BytesIO(raw_bytes)).convert("RGB")
except Exception as exc:
raise HTTPException(
status_code=400,
detail=f"Geçersiz base64 görüntü verisi: {exc}",
) from exc
if img is None and not raw_md:
sample = synthesize_demo_document(doc_type=req.doc_type, apply_degradation=True, seed=2026)
img = sample.image
raw_md = inject_realistic_ocr_corruptions(sample)
elif img is None:
img = Image.new("RGB", (900, 600), color=(255, 255, 255))
page_out = pipeline.postprocess_document_markdown(
raw_markdown=raw_md,
image=img,
repair_financials=True,
)
redaction_res = visual_redactor.redact_document(
image=img,
raw_text=page_out.raw_markdown,
normalized_text=page_out.markdown,
)
buf = io.BytesIO()
redaction_res.redacted_image.save(buf, format="PNG")
redacted_b64 = base64.b64encode(buf.getvalue()).decode("ascii")
return {
"brand": "TürkçeOCR",
"raw_markdown": page_out.raw_markdown,
"verified_markdown": page_out.markdown,
"redacted_markdown": redaction_res.redacted_text,
"entity_count": len(redaction_res.entities),
"entities": [e.to_dict() for e in redaction_res.entities],
"redaction_box_count": len(redaction_res.redaction_boxes),
"redaction_boxes": [b.to_dict() for b in redaction_res.redaction_boxes],
"redaction_coverage": round(float(redaction_res.redaction_coverage), 4),
"redacted_image_base64": redacted_b64,
}
return api
app = create_turkceocr_app()
def main() -> None:
"""`turkceocr-serve` konsol sunucu başlatıcısı."""
import uvicorn
parser = argparse.ArgumentParser(description="TürkçeOCR FastAPI Sunucusu")
parser.add_argument("--host", type=str, default="127.0.0.1")
parser.add_argument("--port", type=int, default=8000)
args = parser.parse_args()
uvicorn.run(app, host=args.host, port=args.port)
if __name__ == "__main__":
main()
|