app.py

import gradio as gr
import tensorflow as tf
import numpy as np
from PIL import Image
import cv2
import datetime
from tensorflow.keras import backend as K

# Define the custom FixedDropout layer
class FixedDropout(tf.keras.layers.Layer):
    def __init__(self, rate, noise_shape=None, seed=None, **kwargs):
        super(FixedDropout, self).__init__(**kwargs)
        self.rate = rate
        self.noise_shape = noise_shape  # Include the noise_shape argument
        self.seed = seed  # Include the seed argument

    def call(self, inputs, training=None):
        if training is None:
            training = K.learning_phase()

        if training:
            return K.in_train_phase(K.dropout(inputs, self.rate, noise_shape=self.noise_shape, seed=self.seed), inputs, training=training)
        else:
            return inputs

    def get_config(self):
        config = super(FixedDropout, self).get_config()
        config['rate'] = self.rate  # Serialize the rate argument
        config['noise_shape'] = self.noise_shape  # Serialize the noise_shape argument
        config['seed'] = self.seed  # Serialize the seed argument
        return config

class ImageClassifierCataract:
    def __init__(self, model_path):
        self.model_path = model_path
        self.model = self.load_model()
        self.class_labels = ["Normal", "Cataract"]

    def load_model(self):
        # Load the trained TensorFlow model
        with tf.keras.utils.custom_object_scope({'FixedDropout': FixedDropout}):
            model = tf.keras.models.load_model(self.model_path)
        return model

    def classify_image(self, input_image):
        input_image = tf.image.resize(input_image, (192, 256))
        input_image = (input_image / 255.0)
        input_image = np.expand_dims(input_image, axis=0)

        current_time = datetime.datetime.now()
        prediction = self.model.predict(input_image)
        class_index = np.argmax(prediction)
        predicted_class = self.class_labels[class_index]

        output_image = (input_image[0] * 255).astype('uint8')
        output_image = cv2.copyMakeBorder(output_image, 0, 50, 0, 0, cv2.BORDER_CONSTANT, value=(255, 255, 255))
        label_background = np.ones((50, output_image.shape[1], 3), dtype=np.uint8) * 255
        output_image[-50:] = label_background

        image_height, image_width, _ = output_image.shape
        box_size = 100
        box_x = (image_width - box_size) // 2
        box_y = (image_height - box_size) // 2
        object_box_color = (255, 0, 0)
        cv2.rectangle(output_image, (box_x, box_y), (box_x + box_size, box_y + box_size), object_box_color, 2)

        # Create a formatted HTML block with the provided information in Portuguese
        info_html = """
        <div style="background-color: #f2f2f2; padding: 10px;">
            <h2>Detecção de Catarata</h2>
            <p>A catarata é uma das doenças oculares mais graves que pode causar cegueira se não for tratada. A detecção da doença em estágios iniciais, em vez de estágios avançados, pode evitar que o paciente fique cego.</p>
            <p>Neste ponto, os pacientes suspeitos devem ser constantemente examinados. O controle contínuo e o acompanhamento dos pacientes são processos cansativos e laboriosos. Por essas razões, neste artigo, são propostos dois modelos de aprendizado profundo diferentes que podem ser usados no diagnóstico e detecção de catarata para auxiliar o trabalho e os procedimentos dos oftalmologistas.</p>
            <p>Os modelos de aprendizado profundo propostos foram executados em um conjunto de dados de fundo de olho com sintomas normais e de catarata.</p>
        </div>
        """

        # Combine the output elements
        output = {
            "image": output_image,
            "text": f"Predicted Class: {predicted_class}",
            "html": info_html,
        }

        return output

    def run_interface(self):
        input_interface = gr.Interface(
            fn=self.classify_image,
            inputs="image",
            outputs=["image", "text", "html"],
            live=True
        )
        input_interface.launch()

if __name__ == "__main__":
    model_path = 'modelo_treinado.h5'  # Replace with the path to your trained model
    app = ImageClassifierCataract(model_path)
    app.run_interface()