Update mdr_pdf_parser.py

mdr_pdf_parser.py

@@ -1125,6 +1125,17 @@ class _MDR_TextRecognizer(_MDR_PredictBase):
     padding = np.zeros((imgC, imgH, imgW), dtype=np.float32)
     padding[:, :, 0:tw] = resized # Place resized image into padded canvas
     print(f"    DEBUG RECOGNIZER: _resize_norm output padded shape: {padding.shape}")
+     # ---- START LOGGING NORMALIZED CROP PROPERTIES ----
+    print(f"    DEBUG RECOGNIZER: Normalized Crop Properties (before ONNX): "
+          f"dtype: {padding.dtype}, " # Should be float32
+          f"MinPx: {np.min(padding):.4f}, "
+          f"MaxPx: {np.max(padding):.4f}, "
+          f"MeanPx: {np.mean(padding):.4f}")
+    if np.all(padding == 0):
+        print("    DEBUG RECOGNIZER: WARNING - Normalized image is all zeros!")
+    elif np.all(padding == padding[0,0,0]): # Check if all elements are the same
+         print(f"    DEBUG RECOGNIZER: WARNING - Normalized image is a constant value: {padding[0,0,0]}")
+    # ---- END LOGGING NORMALIZED CROP PROPERTIES ----
     return padding
 
   def __call__(self, img_list):
@@ -1218,6 +1229,26 @@ class _MDR_TextSystem:
     boxes_for_valid_crops = [boxes[i] for i in valid_idxs]
     print(f"  DEBUG OCR SYS: Number of valid crops to process: {len(valid_crops)}") # DEBUG
 
+    # ---- START LOGGING CROP PROPERTIES ----
+    if valid_crops:
+        print("  DEBUG OCR SYS: Logging properties of first few valid crops (and Box 21 if present):")
+        indices_to_log = list(range(min(3, len(valid_crops)))) # Log first 3
+        # Try to find original index of Box 21 if we can map it back, this is a bit tricky here
+        # For simplicity, let's just log the first few. If Box 21 was among them, we'd see it.
+    
+        for i_log_idx, crop_idx in enumerate(indices_to_log):
+            crop_image_np = valid_crops[crop_idx]
+            if crop_image_np is not None and crop_image_np.size > 0:
+                print(f"    Crop for Recognizer (Index {crop_idx}): "
+                      f"Shape: {crop_image_np.shape}, "
+                      f"dtype: {crop_image_np.dtype}, "
+                      f"MinPx: {np.min(crop_image_np)}, "
+                      f"MaxPx: {np.max(crop_image_np)}, "
+                      f"MeanPx: {np.mean(crop_image_np):.2f}")
+            else:
+                print(f"    Crop for Recognizer (Index {crop_idx}): Is None or empty.")
+    # ---- END LOGGING CROP PROPERTIES ----
+
     if self.use_cls and self.classifier and classify:
         print(f"  DEBUG OCR SYS: Applying classifier to {len(valid_crops)} crops...") # DEBUG
         try: