app.py

import gradio as gr
import tensorflow as tf
import numpy as np
from PIL import Image
 
model_path = "loretmar_ResNet50.keras"
model = tf.keras.models.load_model(model_path)
 

def predict_snake(image):
    # Preprocess image
    print(type(image))
    image = Image.fromarray(image.astype('uint8'))  # Convert numpy array to PIL image
    image = image.resize((200, 200))
    image = np.array(image)
    image = np.expand_dims(image, axis=0) # same as image[None, ...]

    prediction = model.predict(image)
 
    # No need to apply sigmoid, as the output layer already uses softmax
    # Convert the probabilities to rounded values
    prediction = np.round(prediction, 2)
 
    # Separate the probabilities for each class
    p_1 = prediction[0][0]  
    p_2 = prediction[0][1]   
    p_3 = prediction[0][2]    
    p_4 = prediction[0][3]    
 
    return {'Agkistrodon contortrix (venomous)': p_1, 'Agkistrodon piscivorus (venomous)': p_2, 'Ahaetulla nasuta (not venomous)': p_3, 'Ahaetulla prasina (not venomous)': p_4}
 
 
input_image = gr.Image()
iface = gr.Interface(
    fn=predict_snake,
    inputs=input_image,
    outputs=gr.Label(),
    examples=["train/18/0bd1af4119054513917caa7944efd082.jpg", "train/20/0b103c4331bb47f4890d6e0ec96bf9bf.jpg", "train/25/0b4ef7c358044df1b734ea188811e684.jpg", "train/26/0f13f1c263c94602b0462353ff3d188b.jpg"],
    description="TEST.")
 
iface.launch()