app.py

import gradio as gr
import tensorflow as tf
import numpy as np
from PIL import Image
from huggingface_hub import hf_hub_download
import os
import pandas as pd
import logging

# Disable GPU if not available (for Hugging Face Spaces)
os.environ['CUDA_VISIBLE_DEVICES'] = '-1'

# Setup logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Configuration
MODEL_REPO = "Ahmedhassan54/Image-Classification-Model"
MODEL_FILE = "best_model.keras"

# Initialize model
model = None

def load_model():
    global model
    try:
        logger.info("Downloading model...")
        model_path = hf_hub_download(
            repo_id=MODEL_REPO,
            filename=MODEL_FILE,
            cache_dir=".",
            force_download=True
        )
        logger.info(f"Model path: {model_path}")
        
        # Explicitly disable GPU
        with tf.device('/CPU:0'):
            model = tf.keras.models.load_model(model_path)
        logger.info("Model loaded successfully!")
    except Exception as e:
        logger.error(f"Model loading failed: {str(e)}")
        model = None

# Load model at startup
load_model()

def classify_image(image):
    try:
        if image is None:
            return {"Cat": 0.5, "Dog": 0.5}, pd.DataFrame({'Class': ['Cat', 'Dog'], 'Confidence': [0.5, 0.5]})
            
        # Convert to PIL Image if numpy array
        if isinstance(image, np.ndarray):
            image = Image.fromarray(image.astype('uint8'))
            
        # Preprocess
        image = image.resize((150, 150))
        img_array = np.array(image) / 255.0
        if len(img_array.shape) == 3:
            img_array = np.expand_dims(img_array, axis=0)
            
        # Predict
        if model is not None:
            with tf.device('/CPU:0'):
                pred = model.predict(img_array, verbose=0)
            confidence = float(pred[0][0])
        else:
            confidence = 0.75  # Demo value
            
        results = {
            "Cat": round(1 - confidence, 4),
            "Dog": round(confidence, 4)
        }
        
        plot_data = pd.DataFrame({
            'Class': ['Cat', 'Dog'],
            'Confidence': [1 - confidence, confidence]
        })
        
        return results, plot_data
        
    except Exception as e:
        logger.error(f"Error: {str(e)}")
        return {"Error": str(e)}, pd.DataFrame({'Class': ['Cat', 'Dog'], 'Confidence': [0.5, 0.5]})

# Interface
with gr.Blocks() as demo:
    gr.Markdown("# 🐾 Cat vs Dog Classifier 🦮")
    
    with gr.Row():
        with gr.Column():
            img_input = gr.Image(type="pil")
            classify_btn = gr.Button("Classify", variant="primary")
        
        with gr.Column():
            label_out = gr.Label(num_top_classes=2)
            plot_out = gr.BarPlot(
                pd.DataFrame({'Class': ['Cat', 'Dog'], 'Confidence': [0.5, 0.5]}),
                x="Class", y="Confidence", y_lim=[0,1]
            )
    
    classify_btn.click(
        classify_image,
        inputs=img_input,
        outputs=[label_out, plot_out]
    )

    # Examples section
    gr.Examples(
        examples=[
            ["https://upload.wikimedia.org/wikipedia/commons/1/15/Cat_August_2010-4.jpg"],
            ["https://upload.wikimedia.org/wikipedia/commons/d/d9/Collage_of_Nine_Dogs.jpg"]
        ],
        inputs=img_input,
        outputs=[label_out, plot_out],
        fn=classify_image,
        cache_examples=True
    )

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=7860)