Update utils/ocr_utils.py

utils/ocr_utils.py

@@ -41,29 +41,33 @@ def group_nearby_boxes(lines, max_y_gap=50):
 def extract_and_translate_chunk(image: Image.Image):
     np_img = np.array(image)
     results = ocr_model.ocr(np_img)
-    print(results)
+
+    if not results or not isinstance(results[0], dict):
+        print("⚠️ No OCR results or unexpected format")
+        return []
+
+    ocr_data = results[0]
+
+    rec_texts = ocr_data.get("rec_texts", [])
+    rec_polys = ocr_data.get("rec_polys", [])
+
+    if not rec_texts or not rec_polys:
+        print("⚠️ Missing OCR text or polygons")
+        return []
+
+    # Step 1: Build list of (polygon, text)
     lines = []
-    for entry in results[0]:
-        box_raw = entry[0]
-        text_info = entry[1]
-    
-        if isinstance(text_info, (tuple, list)):
-            text = text_info[0]
-        else:
-            text = ""
-    
-        if not text.strip():
-            continue
-    
-        if all(isinstance(pt, (list, tuple)) and len(pt) == 2 for pt in box_raw):
-            polygon = [(int(x), int(y)) for x, y in box_raw]
-            lines.append((polygon, text))  # ✅ This creates the proper format
+    for poly, text in zip(rec_polys, rec_texts):
+        polygon = [(int(x), int(y)) for x, y in poly]
+        lines.append((polygon, text.strip()))
 
     print("🔍 OCR Raw Output:", lines)
 
+    # Step 2: Group nearby text boxes
     grouped = group_nearby_boxes(lines)
-    translations = []
 
+    # Step 3: Translate and compute convex hulls
+    translations = []
     for group in grouped:
         polygons = group["polygons"]
         merged_text = "".join(group["texts"]).strip()
@@ -77,10 +81,9 @@ def extract_and_translate_chunk(image: Image.Image):
             print("⚠️ Translation failed:", e)
             translated = ""
 
-        # Flatten all polygon points for convex hull
         all_points = np.array([pt for polygon in polygons for pt in polygon])
         if len(all_points) < 3:
-            continue  # convex hull needs at least 3 points
+            continue  # convex hull needs ≥ 3
 
         hull_indices = ConvexHull(all_points).vertices
         hull = [tuple(map(int, all_points[i])) for i in hull_indices]