Update app.py

app.py

@@ -1,26 +1,67 @@
+import gradio as gr
+from paddleocr import PaddleOCR
+from PIL import Image
+import numpy as np
+import logging
+
+# Disable internal logging to prevent "Invalid File Descriptor" noise
+logging.getLogger("ppocr").setLevel(logging.ERROR)
+
+# 1. BARE MINIMAL INITIALIZATION
+# In 3.0+, just set the version and language. 
+# Angle classification is handled internally.
+try:
+    ocr = PaddleOCR(
+        lang='en', 
+        ocr_version='PP-OCRv4',
+        use_angle_cls=True
+    )
+except Exception as e:
+    print(f"Startup Error: {e}")
+
 def process_bank_form(image):
     if image is None:
         return "Please upload an image."
 
-    # Convert PIL to RGB numpy array
-    img = image.convert("RGB")
-    img_array = np.array(img)
-    
-    # NEWEST API (3.0+): 
-    # Do not pass 'cls=True' here. 
-    # It is already handled because we set 'use_angle_cls=True' in the initializer.
-    result = ocr.ocr(img_array)
-
-    if not result or result[0] is None:
-        return "No text detected. Try a clearer photo or a darker pen."
-
-    # Process and combine text
-    extracted_text = []
-    for line in result[0]:
-        text = line[1][0]      # The recognized string
-        confidence = line[1][1] # Confidence score
+    try:
+        # 2. Standardize Image Format
+        # Converting to RGB and then a Numpy array prevents "Invalid Format" errors
+        img = image.convert("RGB")
+        img_array = np.array(img)
         
-        if confidence > 0.4:
-            extracted_text.append(text)
-    
-    return "\n".join(extracted_text)
+        # 3. Safe Inference
+        # Removed 'cls=True' to fix the TypeError you saw earlier.
+        # PaddleOCR 3.0+ handles the pipeline automatically.
+        result = ocr.ocr(img_array)
+
+        if not result or result[0] is None:
+            return "No text detected. Try a closer photo with better lighting."
+
+        # 4. Extract Text
+        extracted_text = []
+        for line in result[0]:
+            # line[1][0] is the text string
+            text_str = str(line[1][0])
+            confidence = float(line[1][1])
+            
+            # Lower confidence threshold slightly for handwriting (0.35)
+            if confidence > 0.35:
+                extracted_text.append(text_str)
+        
+        return "\n".join(extracted_text)
+
+    except Exception as e:
+        return f"Runtime Error: {str(e)}\nTip: Try a smaller image or restart the Space."
+
+# 5. UI Setup
+demo = gr.Interface(
+    fn=process_bank_form,
+    inputs=gr.Image(type="pil", label="Bank Form Image"),
+    outputs=gr.Textbox(label="Extracted Content", lines=20),
+    title="🏦 Global Bank Form OCR",
+    description="Optimized for English handwriting and printed text on CPU."
+)
+
+if __name__ == "__main__":
+    # max_threads=1 is safer for free CPU tiers to prevent memory crashes
+    demo.launch(max_threads=1)