app.py

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

# Load the trained MNIST model
model = tf.keras.models.load_model("./sketch_recognition_numbers_model.h5")

# Class names (0 to 9)
labels = ["zero", "one", "two", "three", "four", "five", "six", "seven", "eight", "nine"]

def predict(data):
    # Extract the 'composite' key from the input dictionary
    img = data["composite"]
    img = np.array(img)

    # Convert RGBA to RGB if needed
    if img.shape[-1] == 4:  # RGBA
        img = cv2.cvtColor(img, cv2.COLOR_RGBA2RGB)

    # Convert RGB to Grayscale
    if img.shape[-1] == 3:  # RGB
        img = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)

    # Resize image to 28x28
    img = cv2.resize(img, (28, 28))

    # Normalize pixel values to [0, 1]
    img = img / 255.0

    # Reshape to match model input (1, 28, 28, 1)
    img = img.reshape(1, 28, 28, 1)

    # Model predictions
    preds = model.predict(img)[0]

    print(preds)

    # Get top 3 classes
    top_3_classes = np.argsort(preds)[-3:][::-1]
    top_3_probs = preds[top_3_classes]
    class_names = [labels[i] for i in top_3_classes]
    print(class_names, top_3_probs, top_3_classes)

    # Return top 3 predictions as a dictionary
    return {class_names[i]: float(top_3_probs[i]) for i in range(3)}

# Title and description
title = "Welcome to your first sketch recognition app!"
head = (
    "<center>"
    "<img src='./mnist-classes.png' width=400>"
    "<p>The model is trained to classify numbers (from 0 to 9). "
    "To test it, draw your number in the space provided (use the editing tools in the image editor).</p>"
    "</center>"
)
ref = "Find the complete code [here](https://github.com/ovh/ai-training-examples/tree/main/apps/gradio/sketch-recognition)."

with gr.Blocks(title=title) as demo:
    # Display title and description
    gr.Markdown(head)
    gr.Markdown(ref)

    with gr.Row():
        # Using ImageEditor with type='numpy'
        im = gr.Sketchpad(type="numpy", label="Draw your digit here (use brush and eraser)")

        # Output label (top 3 predictions)
        label = gr.Label(num_top_classes=3, label="Predictions")

    # Trigger prediction whenever the image changes
    im.change(predict, inputs=im, outputs=label)

    demo.launch(share=True)