app.py

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

# App title
title = "Welcome to your first sketch recognition app!"

# App description
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.</p>"
    "</center>"
)

# GitHub repository link
ref = "Find the complete code [here](https://github.com/ovh/ai-training-examples/tree/main/apps/gradio/sketch-recognition)."


# Class names (from 0 to 9)
labels = {
    0: "zero", 
    1: "one", 
    2: "two",
    3: "three",
    4: "four",
    5: "five",
    6: "six",
    7: "seven",
    8: "eight",
    9: "nine"    
}
# Load model (trained on MNIST dataset)
model = tf.keras.models.load_model("./sketch_recognition_numbers_model.h5")

def predict(data):
    # Convert to NumPy array
    img = np.array(data['composite'])

    print("img.shape", img.shape)

    # Handle RGBA or RGB images
    if img.shape[-1] == 4:  # RGBA
        img = cv2.cvtColor(img, cv2.COLOR_RGBA2RGB)
    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
    img = img.reshape(1, 28, 28, 1)

    print("img", img)

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

    print("preds", preds)
    values_map = {preds[i]: i for i in range(len(preds))}
    
    sorted_values = sorted(preds, reverse=True)

    labels_map = dict()
    for i in range(3):
        print("sorted_values[i]", sorted_values[i], values_map[sorted_values[i]])
        labels_map[labels[values_map[sorted_values[i]]]] = sorted_values[i]

    print("labels_map", labels_map)
    return labels_map

# Top 3 classes
label = gr.Label(num_top_classes=3)

# Open Gradio interface for sketch recognition
interface = gr.Interface(
    fn=predict,
    inputs=gr.Sketchpad(type='numpy'),
    outputs=label,
    title=title,
    description=head,
    article=ref
)
interface.launch(share=True)