Update app.py

app.py

@@ -22,54 +22,69 @@ labels = ["zero", "one", "two", "three", "four", "five", "six", "seven", "eight"
 
 def predict(img):
     try:
-        # Check if the image is a PIL object and convert it to NumPy array
+        # Enhanced image validation and conversion
+        if img is None:
+            raise ValueError("No image provided")
+        
+        # Convert to numpy array if it's a PIL Image
         if isinstance(img, Image.Image):
             img = np.array(img)
-
-        # If the image is base64, convert it to a PIL image and then to a NumPy array
-        elif isinstance(img, str):  # base64 image
-            img = Image.open(io.BytesIO(base64.b64decode(img.split(',')[1])))
-            img = np.array(img)
-
-        # If the input is still not a NumPy array, raise an error
+        
+        # Handle base64 image strings
+        elif isinstance(img, str):
+            # Check if it's a base64 data URL
+            if img.startswith('data:image'):
+                # Split and decode base64 part
+                img = img.split(',')[1]
+            
+            # Decode base64 to image
+            try:
+                img = Image.open(io.BytesIO(base64.b64decode(img)))
+                img = np.array(img)
+            except Exception as e:
+                print(f"Base64 decoding error: {e}")
+                raise ValueError("Invalid base64 image")
+        
+        # Validate numpy array
         if not isinstance(img, np.ndarray):
-            raise ValueError("Input is not a valid image")
-
-        # Print shape and type of the input image
+            raise ValueError("Input could not be converted to a valid image")
+        
+        # Print initial image details for debugging
         print(f"Initial image type: {type(img)}, shape: {img.shape}")
-
-        # Ensure the image is in grayscale and has a single channel
-        if img.ndim == 3 and img.shape[-1] == 3:
-            img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-        elif img.ndim == 2:
+        
+        # Handle color channels
+        if img.ndim == 3:
+            if img.shape[-1] == 3:  # Color image
+                img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+            elif img.shape[-1] == 4:  # RGBA image
+                img = cv2.cvtColor(img, cv2.COLOR_RGBA2GRAY)
+        
+        # Ensure single channel
+        if img.ndim == 2:
             img = np.expand_dims(img, axis=-1)
-
-        # Print the shape of the grayscale image
-        print(f"Grayscale image shape: {img.shape}")
-
-        # Resize the image
+        
+        # Resize and normalize
         img = cv2.resize(img, (img_size, img_size))
-
-        # Normalize the image
         img = img.astype('float32') / 255.0
         img = img.reshape(1, img_size, img_size, 1)
-
-        # Print the shape after resizing and normalizing
+        
+        # Print processed image details
         print(f"Processed image shape: {img.shape}")
-
-        # Get the predictions from the model
+        
+        # Get predictions from the model
         preds = model.predict(img)[0]
-
-        # Print the predictions
+        
+        # Print predictions for debugging
         print("Predictions:", preds)
-
-        # Return the predictions for each label
+        
+        # Return predictions as a dictionary
         return {label: float(pred) for label, pred in zip(labels, preds)}
+    
     except Exception as e:
-        # Print the exception to the console
-        print(f"Error during prediction: {e}")
+        # Comprehensive error logging
+        print(f"Full error during prediction: {e}")
         return {"Error": str(e)}
-
+            
    
 
 # Set up the Gradio interface with the input as a sketchpad and output as labels
@@ -78,8 +93,8 @@ label = gr.Label(num_top_classes=3)
 # Gradio arayüzü
 interface = gr.Interface(
     fn=predict,
-    inputs="sketchpad",
-    outputs="textbox",
+    inputs=gr.Sketchpad(type="pil"),
+    outputs=gr.Label(num_top_classes=3), ,
     title="Sketch Recognition App",
     description="Draw a number (0-9) and see the model's top predictions."
 )