import gradio as gr
import tensorflow as tf
from PIL import Image
import numpy as np
import cv2

# Loading saved model
model = tf.keras.models.load_model('gender_recognition.h5')

def predict(input_image):
    try:
        # Convert PIL Image to OpenCV format (numpy array)
        input_image = cv2.cvtColor(np.array(input_image), cv2.COLOR_RGB2BGR)
        # Resizing and preprocessing input image
        input_image = cv2.resize(input_image, (178, 218))
        input_image = np.array(input_image).astype(np.float32) / 255.0
        input_image = np.expand_dims(input_image, axis=0)  # Add a dimension for the batch size

        # Making prediction
        prediction = model.predict(input_image)

        # Postprocess prediction
        labels = ['Female', 'Male']
        threshold = 0.5  # threshold for classifying as 'Male'
        predicted_gender = 'Male' if prediction[0][1] > threshold else 'Female'
        prediction_probability = prediction[0][1] if predicted_gender == 'Male' else prediction[0][0]

        male_emoji = "\U0001F468"  # Man emoji
        female_emoji = "\U0001F469"  # Woman emoji

        selected_emoji = male_emoji if predicted_gender == 'Male' else female_emoji

        # Combine the predicted gender and the probability into a single string
        output = f"{selected_emoji} {predicted_gender}\n{prediction_probability * 100:.2f}% probability."

        return output
    except Exception as e:
        return str(e)


# Creating Gradio interface
iface = gr.Interface(
    fn=predict, 
    inputs=gr.inputs.Image(shape=(218, 178)), 
    outputs="text",
    title = 'Image Recognition - Gender Detection with InceptionV3',
    description="""<br> This model was trained to predict the gender of a person based on a photo. <br> 
    The training of this model can be seen on this <a href='https://www.kaggle.com/code/lusfernandotorres/gender-recognition-inceptionv3'>Kaggle notebook</a>. <br>
    <br>Upload a photo to see the how the model predicts the gender of the person on it!"""
)

iface.launch()