feat add glm-ocr core

app/services/layout_detector.py

@@ -1,9 +1,10 @@
-"""PP-DocLayoutV2 wrapper for document layout detection."""
+"""PP-DocLayoutV3 wrapper for document layout detection."""
 
 import numpy as np
 
 from app.schemas.image import LayoutInfo, LayoutRegion
 from app.core.config import get_settings
+from app.services.layout_postprocess import apply_layout_postprocess
 from paddleocr import LayoutDetection
 from typing import Optional
 
@@ -116,6 +117,17 @@ class LayoutDetector:
         else:
             boxes = []
 
+        # Apply GLM-OCR layout post-processing (NMS, containment, unclip, clamp)
+        if boxes:
+            h, w = image.shape[:2]
+            boxes = apply_layout_postprocess(
+                boxes,
+                img_size=(w, h),
+                layout_nms=True,
+                layout_unclip_ratio=None,
+                layout_merge_bboxes_mode="large",
+            )
+
         for box in boxes:
             cls_id = box.get("cls_id")
             label = box.get("label") or self.CLS_ID_TO_LABEL.get(cls_id, "other")
@@ -128,6 +140,7 @@ class LayoutDetector:
             regions.append(
                 LayoutRegion(
                     type=region_type,
+                    native_label=label,
                     bbox=coordinate,
                     confidence=score,
                     score=score,