app.py

import tensorflow as tf
import gradio as gr
import cv2
import numpy as np

# 1. Load your saved model
model = tf.keras.models.load_model('digit_recognizer.keras')

def classify_digit(image):
    # Error handling: if no image is provided
    if image is None:
        return None

    # --- PREPROCESSING ---
    # Convert to numpy array if it isn't already
    image = np.array(image)

    # 1. Handle Color Channels
    # If image has 4 channels (RGBA) from sketchpad, convert to Gray
    if image.shape[-1] == 4:
        image = cv2.cvtColor(image, cv2.COLOR_RGBA2GRAY)
    # If image has 3 channels (RGB) from upload, convert to Gray
    elif image.shape[-1] == 3:
        image = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
    
    # 2. Resize to 28x28
    # We use INTER_AREA for shrinking which preserves details better than default
    image = cv2.resize(image, (28, 28), interpolation=cv2.INTER_AREA)

    # 3. Invert Colors (Critical Step)
    # MNIST models expect White Text on Black Background.
    # If the image is mostly bright (like white paper), we must invert it.
    avg_brightness = np.mean(image)
    if avg_brightness > 127:  # If the image is mostly white/light
        image = 255 - image   # Invert to black background

    # 4. Reshape for Model
    # (1 sample, 28 height, 28 width, 1 channel)
    image = image.reshape(1, 28, 28, 1)
    
    # 5. Normalize (0 to 1)
    image = image / 255.0

    # --- PREDICTION ---
    prediction = model.predict(image).flatten()
    return {str(i): float(prediction[i]) for i in range(10)}

# --- GRADIO INTERFACE ---
# sources=["upload", "canvas"] enables both file upload and drawing
interface = gr.Interface(
    fn=classify_digit,
    inputs=gr.Image(
        type="numpy", 
        label="Draw or Upload Digit", 
        image_mode="L",  # "L" attempts to convert to grayscale immediately
        sources=["upload", "canvas"],
        height=400,
        width=400
    ),
    outputs=gr.Label(num_top_classes=3),
    title="Handwritten Digit Recognizer",
    description="Draw a digit on the canvas OR upload a photo of a digit. The model will guess what it is."
)

if __name__ == "__main__":
    interface.launch()