app.py

import gradio as gr
import tensorflow as tf
from transformers import GPT2LMHeadModel, GPT2Tokenizer
import numpy as np
from keras.preprocessing import image
from tensorflow.image import resize
from keras.layers import BatchNormalization

# Load the image classification model
custom_objects = {'BatchNormalization': BatchNormalization}

# Load the model with custom objects
try:
    model = tf.keras.models.load_model('model1025_ver1.h5', custom_objects=custom_objects)
except Exception as e:
    print(f"An error occurred: {e}")

# Load GPT-2 tokenizer and model
tokenizer = GPT2Tokenizer.from_pretrained("diamantrsd/copywriting-otomatis")
gpt2_model = GPT2LMHeadModel.from_pretrained("diamantrsd/copywriting-otomatis")

def predict_image(img_path, keyword=""):
    if img_path is None:
        return "Error: Belum upload gambar. Silakan upload gambar"
    # Preprocess the image
    img = image.load_img(img_path, target_size=(224, 224))
    img_array = image.img_to_array(img)
    img_array = np.expand_dims(img_array, axis=0)
    img_array = img_array / 255.0

    # Perform inference with the model
    predictions = model.predict(img_array)

    # Get the predicted class
    predicted_class = np.argmax(predictions[0])

    # You might need to create a dictionary mapping class indices to their labels
    class_labels = {
        0: 'Backpack',
        1: 'Celana Panjang',
        2: 'Celana Pendek',
        3: 'Dompet',
        4: 'Dress',
        5: 'kacamata',
        6: 'Kaos',
        7: 'Kaos Kaki',
        8: 'Kemeja',
        9: 'Outerwear',
        10: 'Sandal',
        11: 'Sepatu',
        12: 'Sepatu Flat',
        13: 'Tas',
        14: 'Topi'
    }

    # Get class label
    class_label = class_labels[predicted_class]

    # Generate copywriting based on class label
    copywriting = generate_copywriting(class_label, keyword)

    # Return the prediction result and copywriting
    return class_label, copywriting

# Function to generate copywriting
def generate_copywriting(class_label,keyword):
    # Generate prompt based on class label
    prompt = f"{keyword.lower()}, {class_label.lower()}" if keyword else f"{class_label.lower()}"

    input_ids = tokenizer.encode(prompt, return_tensors="pt")

    # Adjust parameters as needed
    max_length = 100
    no_repeat_ngram_size = 3
    top_k = 35
    top_p = 0.50

    output = gpt2_model.generate(
        input_ids,
        max_length=max_length,
        no_repeat_ngram_size=no_repeat_ngram_size,
        top_k=top_k,
        top_p=top_p,
        pad_token_id=tokenizer.eos_token_id,
        num_return_sequences=1
    )

    generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
    generated_text = generated_text.replace(keyword, '', 1).strip()
    return generated_text

image_input = gr.Image(label="Upload Gambar Fashion Kamu", type="filepath")
keyword_input = gr.Textbox(label="Keyword Tambahan (Opsional)")
output_label = gr.Textbox(label="Kelas Prediksi")
output_copywriting = gr.Textbox(label="Copywriting")

# Add the prediction function to the Gradio interface
iface = gr.Interface(fn=predict_image, inputs=[image_input, keyword_input], outputs=[output_label, output_copywriting], title="Copywriting Otomatis")


# Run the Gradio interface
iface.launch()