feat(ocr): bump to latest surya 0.17.1, kraken 7.0, torch 2.7.1

- surya-ocr 0.6.3 → 0.17.1: new predictor API (FoundationPredictor,
  RecognitionPredictor, DetectionPredictor), native polygon output
  on text lines (4-point clockwise)
- kraken 5.2.9 → 7.0: wider torch range (>=2.4,<=2.10), unpinned numpy
- torch 2.5.1 → 2.7.1: satisfies surya's >=2.7.0 requirement
- Rewrite engines/surya.py for the 0.17 predictor class API
- Surya now outputs polygons natively — no longer rectangle-only

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

ocr-service/Dockerfile

@@ -10,7 +10,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 
 # PyTorch CPU-only — separate layer; the whl/cpu index strips all CUDA variants (~2 GB saved)
 RUN pip install --no-cache-dir \
-    torch==2.5.1 \
+    torch==2.7.1 \
     --index-url https://download.pytorch.org/whl/cpu
 
 COPY requirements.txt .

ocr-service/engines/surya.py

@@ -4,28 +4,23 @@ import logging
 
 logger = logging.getLogger(__name__)
 
-# Lazy-loaded at startup via load_models()
+_recognition_predictor = None
-_recognition_model = None
+_detection_predictor = None
-_recognition_processor = None
-_detection_model = None
-_detection_processor = None
 
 
 def load_models():
     """Eagerly load Surya models into memory. Called once at container startup."""
-    global _recognition_model, _recognition_processor, _detection_model, _detection_processor
+    global _recognition_predictor, _detection_predictor
 
     logger.info("Loading Surya models...")
 
-    from surya.model.detection.model import load_model as load_det_model
+    from surya.foundation import FoundationPredictor
-    from surya.model.detection.model import load_processor as load_det_processor
+    from surya.recognition import RecognitionPredictor
-    from surya.model.recognition.model import load_model as load_rec_model
+    from surya.detection import DetectionPredictor
-    from surya.model.recognition.processor import load_processor as load_rec_processor
 
-    _detection_model = load_det_model()
+    foundation_predictor = FoundationPredictor()
-    _detection_processor = load_det_processor()
+    _recognition_predictor = RecognitionPredictor(foundation_predictor)
-    _recognition_model = load_rec_model()
+    _detection_predictor = DetectionPredictor()
-    _recognition_processor = load_rec_processor()
 
     logger.info("Surya models loaded successfully")
 
@@ -33,33 +28,36 @@ def load_models():
 def extract_blocks(images: list, language: str = "de") -> list[dict]:
     """Run Surya OCR on a list of PIL images (one per page).
 
-    Returns a flat list of block dicts with pageNumber, x, y, width, height, text.
+    Returns a flat list of block dicts with pageNumber, x, y, width, height,
-    Coordinates are normalized to [0, 1] relative to page dimensions.
+    polygon, text. Coordinates are normalized to [0, 1] relative to page dimensions.
+    Surya 0.17+ returns polygon (4-point) natively on each text line.
     """
-    from surya.detection import batch_text_detection
-    from surya.recognition import batch_recognition
-
     all_blocks = []
 
-    for page_idx, image in enumerate(images):
+    predictions = _recognition_predictor(images, det_predictor=_detection_predictor)
-        page_w, page_h = image.size
 
-        det_predictions = batch_text_detection([image], _detection_model, _detection_processor)
+    for page_idx, page_pred in enumerate(predictions):
-        rec_predictions = batch_recognition(
+        page_w, page_h = images[page_idx].size
-            [image], det_predictions, _recognition_model, _recognition_processor, [language]
-        )
 
-        for line in rec_predictions[0].text_lines:
+        for line in page_pred.text_lines:
             bbox = line.bbox  # [x1, y1, x2, y2] in pixel coordinates
             x1, y1, x2, y2 = bbox
 
+            # Surya 0.17 provides polygon as list of (x, y) tuples (4 points, clockwise)
+            polygon = None
+            if hasattr(line, "polygon") and line.polygon and len(line.polygon) == 4:
+                polygon = [
+                    [p[0] / page_w, p[1] / page_h]
+                    for p in line.polygon
+                ]
+
             all_blocks.append({
                 "pageNumber": page_idx,
                 "x": x1 / page_w,
                 "y": y1 / page_h,
                 "width": (x2 - x1) / page_w,
                 "height": (y2 - y1) / page_h,
-                "polygon": None,
+                "polygon": polygon,
                 "text": line.text,
             })
 

ocr-service/requirements.txt

@@ -1,6 +1,6 @@
 fastapi[standard]==0.115.6
-surya-ocr==0.6.3
+surya-ocr==0.17.1
-kraken==6.0.3
+kraken==7.0
 pillow>=10.2.0,<11.0.0
 pypdfium2==4.30.0
 httpx==0.28.1