Upload 3 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+digit_recognizer_model.keras filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,80 @@
+import gradio as gr
+import tensorflow as tf
+import numpy as np
+from PIL import Image, ImageOps
+
+# --- 1. Load the Saved Model ---
+try:
+    model = tf.keras.models.load_model('digit_recognizer_model.keras')
+except Exception as e:
+    print(f"Error loading model: {e}")
+    model = None
+
+# --- 2. Create the Prediction Function ---
+def predict_digit(sketchpad_data):
+    if model is None or sketchpad_data is None:
+        return "Model not loaded or no drawing provided."
+
+    # Use the 'composite' key to get the final image with the drawing
+    composite_image = sketchpad_data['composite']
+    
+    if composite_image is None:
+        return "Please draw a digit."
+
+    # --- ADVANCED PREPROCESSING ---
+    
+    # 1. Convert the NumPy array to a Pillow Image object, letting Pillow detect the mode
+    pil_image = Image.fromarray(composite_image.astype('uint8')).convert('L')
+    
+    # 2. Invert colors (so digit is white, background is black)
+    inverted_image = ImageOps.invert(pil_image)
+    
+    # 3. Find the bounding box of the digit
+    bbox = inverted_image.getbbox()
+    if bbox is None:
+        return "Please draw a digit." # Handle blank image
+        
+    # 4. Crop the image to the bounding box
+    cropped_image = inverted_image.crop(bbox)
+    
+    # 5. Resize the digit to fit within a 20x20 box to add padding, preserving aspect ratio
+    cropped_image.thumbnail((20, 20), Image.Resampling.LANCZOS)
+    
+    # 6. Create a new 28x28 black canvas and paste the digit in the center
+    new_canvas = Image.new('L', (28, 28), 0) # Black background
+    paste_x = (28 - cropped_image.width) // 2
+    paste_y = (28 - cropped_image.height) // 2
+    new_canvas.paste(cropped_image, (paste_x, paste_y))
+    
+    # 7. Convert the final canvas to a NumPy array for the model
+    image = np.array(new_canvas)
+
+    # 8. Reshape for the model and normalize.
+    image = image.reshape(1, 28, 28).astype('float32') / 255.0
+
+    # --- End of Preprocessing ---
+
+    # Make a prediction
+    prediction = model.predict(image)
+
+    # Format the prediction into a dictionary of confidences
+    confidences = {str(i): float(prediction[0][i]) for i in range(10)}
+    
+    return confidences
+
+# --- 3. Create and Launch the Gradio Interface ---
+inputs = gr.Sketchpad(image_mode='L')
+outputs = gr.Label(num_top_classes=1, label="Prediction")
+
+iface = gr.Interface(
+    fn=predict_digit,
+    inputs=inputs,
+    outputs=outputs,
+    title="Handwritten Digit Recognizer ✍️",
+    description="Draw a digit from 0 to 9 on the canvas, and the model will predict what it is. For best results, draw a single, clear digit.",
+    live=True
+)
+
+# Launch the app!
+print("Launching the Gradio app...")
+iface.launch()

digit_recognizer_model.keras

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a302a3844d78f2f7d90c6621668677ef6e8cc3ad766db750fafd6dc179335d32
+size 8359737

requirements.txt

@@ -0,0 +1,3 @@
+tensorflow
+gradio
+Pillow