app.py

import gradio as gr
from PIL import Image
import numpy as np
import tensorflow as tf
import os

# Configurar variables de entorno para reducir advertencias
os.environ['TF_ENABLE_ONEDNN_OPTS'] = '0'
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

# Configuración inicial
cifar10_labels = [
    'avión', 'automóvil', 'pájaro', 'gato', 'venado',
    'perro', 'rana', 'caballo', 'barco', 'camión'
]

model = tf.keras.models.load_model('my_model.keras')

def preprocess_image(image):
    """Preprocesado de imagen para el modelo"""
    img = image.resize((32, 32)).convert('RGB')
    return np.array(img).astype('float32') / 255

def predict(image):
    """Realizar predicción y formatear resultados"""
    if image is None:
        raise gr.Error("¡Por favor sube una imagen o toma una foto!")
    
    processed_img = preprocess_image(image)
    preds = model.predict(np.expand_dims(processed_img, axis=0))[0]
    return {label: float(preds[i]) for i, label in enumerate(cifar10_labels)}

# Configurar ejemplos
examples = [
    ["ejemplos/avion.jpg"],
    ["ejemplos/automovil.jpg"],
    ["ejemplos/pajaro.jpg"],
    ["ejemplos/gato.jpg"],
    ["ejemplos/venado.jpg"],

]

# Construir interfaz
with gr.Blocks(theme=gr.themes.Soft(), css="""
    .examples-grid {display: flex !important; flex-direction: column; gap: 1rem}
    .examples-row {display: flex !important; gap: 1rem; justify-content: center}
    """) as app:
    
    gr.Markdown("# CIFAR-10con 5 clases")
    
    with gr.Row():
        with gr.Column():
            input_image = gr.Image(
                sources=["upload", "webcam", "clipboard"],
                type="pil",
                label="Entrada de imagen",
                height=250
            )
            with gr.Row():
                submit_btn = gr.Button("Predecir", variant="primary")
                clear_btn = gr.Button("Limpiar")
        
        with gr.Column():
            output_label = gr.Label(
                label="Resultados",
                num_top_classes=3,
                show_label=True
            )
    
    # Sección de ejemplos con interacción
    gr.Markdown("## Ejemplos de categorías")
    with gr.Column(elem_classes=["examples-grid"]):
        # Primera fila
        with gr.Row(elem_classes=["examples-row"]):
            for example, label in zip(examples[:5], cifar10_labels[:5]):
                gr.Examples(
                    examples=example,
                    inputs=[input_image],
                    label=label.capitalize(),
                    examples_per_page=1,
                    fn=predict,
                    outputs=[output_label],
                )
        # Segunda fila
        with gr.Row(elem_classes=["examples-row"]):
            for example, label in zip(examples[5:], cifar10_labels[5:]):
                gr.Examples(
                    examples=example,
                    inputs=[input_image],
                    label=label.capitalize(),
                    examples_per_page=1,
                    fn=predict,
                    outputs=[output_label],
                )

    # Conectar eventos
    submit_btn.click(
        fn=predict,
        inputs=input_image,
        outputs=output_label,
        api_name="predict"
    )
    
    clear_btn.click(
        fn=lambda: [None, None],
        inputs=None,
        outputs=[input_image, output_label]
    )

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