app.py

import keras_nlp
import gradio as gr
import tensorflow as tf
import gdown
from PIL import Image
#import keras_nlp
from tensorflow import keras
import time

input_shape = (32, 32, 3)
resized_shape = (299, 299, 3)
num_classes = 10
labels = {
    0: "plane",
    1: "car",
    2: "bird",
    3: "cat",
    4: "deer",
    5: "dog",
    6: "frog",
    7: "horse",
    8: "ship",
    9: "truck",
}

# Download the NLP model 
def download_model_NLP():
    preprocessor = keras_nlp.models.GPT2CausalLMPreprocessor.from_preset(
    "gpt2_base_en",
    sequence_length=128,)
    model = keras_nlp.models.GPT2CausalLM.from_preset(
    "gpt2_base_en", preprocessor=preprocessor)
    
    output = "total.h5"
    id = "1-KgcnP1ayWQ6l2-4h723JCYPoWxzOnU3"
    gdown.download(id=id, output=output, quiet=False)    
    model.load_weights(output)     
    return model





# Download the model file
def download_model():
    url = "https://drive.google.com/uc?id=1zUGAPg9RVgo7bWtf_-L9MXoXKldZjs1y"
    output = "CIFAR10_Xception_(ACC_0.9704__LOSS_0.0335_).h5"
    gdown.download(url, output, quiet=False)
    return output




model_file = download_model()

# Load the model
model = tf.keras.models.load_model(model_file)

# Perform image classification for single class output
# def predict_class(image):
#     img = tf.cast(image, tf.float32)
#     img = tf.image.resize(img, [input_shape[0], input_shape[1]])
#     img = tf.expand_dims(img, axis=0)
#     prediction = model.predict(img)
#     class_index = tf.argmax(prediction[0]).numpy()
#     predicted_class = labels[class_index]
#     return predicted_class

# Perform image classification for multy class output
def predict_class(image):
    img = tf.cast(image, tf.float32)
    img = tf.image.resize(img, [input_shape[0], input_shape[1]])
    img = tf.expand_dims(img, axis=0)
    prediction = model.predict(img)
    return prediction[0]

# UI Design for single class output
# def classify_image(image):
#     predicted_class = predict_class(image)
#     output = f"<h2>Predicted Class: <span style='text-transform:uppercase';>{predicted_class}</span></h2>"
#     return output


# UI Design for multy class output
def classify_image(image):
    
    results = predict_class(image)
    output = {labels.get(i): float(results[i]) for i in range(len(results))}
    max_value = max(results)
    for i in range(len(results)):
      if results[i] == max_value:
        name = labels.get(i).capitalize()
    
    result_NLP = model_NLP.generate(f"{name} is able to", max_length=100)
    index = result_NLP.find('.', 75)
    result_NLP = result_NLP[0:index+1]
    return output, result_NLP


inputs = gr.inputs.Image(type="pil", label="Upload an image")
# outputs = gr.outputs.HTML() #uncomment for single class output 
output_1 = gr.outputs.Label(num_top_classes=4)

title = "<h1 style='text-align: center;'>Image Classifier with Funny Annotations :)</h1>"
description = "Upload an image and get the predicted class."
# css_code='body{background-image:url("file=wave.mp4");}'

model_NLP  = download_model_NLP()

gr.Interface(fn=classify_image, 
             inputs=inputs, 
             outputs=[output_1, "text"], 
             title=title, 
             examples=[["00_plane.jpg"], ["01_car.jpg"], ["02_bird.jpg"], ["03_cat.jpg"], ["04_deer.jpg"]],
             # css=css_code,
             description=description,
            enable_queue=True).launch()