import gradio as gr
import numpy as np
import os
os.environ['TF_ENABLE_ONEDNN_OPTS'] = '0'
from tensorflow.keras.preprocessing import image
from huggingface_hub import from_pretrained_keras
import requests

# URL of the model file (adjust if needed)
model_url = "https://huggingface.co/diabolic6045/indian_cities_image_classification/resolve/main/model.h5"
model_path = "model.h5"

# Download the model if it doesn't exist
if not os.path.exists(model_path):
    print("Downloading the model...")
    response = requests.get(model_url)
    with open(model_path, "wb") as f:
        f.write(response.content)
    print("Model downloaded.")

from tensorflow.keras.models import load_model
from tensorflow.keras.optimizers import Adam
print("loading model")
# Load the model, ignoring the optimizer argument
model = load_model(model_path, compile=False)

# Recompile the model with a valid optimizer
model.compile(optimizer=Adam(), loss="categorical_crossentropy")

# Define the class labels
class_labels = ['Ahmedabad', 'Delhi', 'Kerala', 'Kolkata', 'Mumbai']

# Function to preprocess the image and predict the city
def classify_city(img):
    # Preprocess the image
    img = img.resize((175, 175))
    img = image.img_to_array(img)
    img = np.expand_dims(img, axis=0)
    img = img / 175.0  # Normalize the image
    
    # Make predictions
    predictions = model.predict(img)
    predicted_class = np.argmax(predictions)
    predicted_city = class_labels[predicted_class]
    
    return f"Predicted City: {predicted_city}"

# Gradio Interface
iface = gr.Interface(
    fn=classify_city, 
    inputs=gr.Image(type="pil", label="Upload an image of an Indian city"), 
    outputs=gr.Textbox(label="Predicted City"),
    title="Indian Cities Image Classification",
    description="Upload an image of a city in India, and the model will predict which city it is: Ahmedabad, Delhi, Kerala, Kolkata, or Mumbai.",
)

# Launch the Gradio app
iface.launch()