app.py

import gradio as gr
from tensorflow.keras.models import load_model
from tensorflow.keras.preprocessing.image import img_to_array
import numpy as np
from PIL import Image

# 🔹 Load your saved model
model = load_model("")

# 🔹 Define your class labels (must match model training)
class_names = ['cardboard', 'glass', 'metal', 'paper', 'plastic', 'trash']

# 🔹 Prediction function
def predict_from_camera(image):
    image = image.resize((224, 224))  # Resize for model input
    img_array = img_to_array(image) / 255.0  # Normalize
    img_array = np.expand_dims(img_array, axis=0)

    prediction = model.predict(img_array)[0]
    predicted_class = class_names[np.argmax(prediction)]
    confidence = float(np.max(prediction))

    return f"{predicted_class} ({confidence*100:.1f}%)"

# 🔹 Gradio live webcam interface
interface = gr.Interface(
    fn=predict_from_camera,
    inputs=gr.Image(source="webcam", streaming=True, type="pil"),
    outputs="text",
    title="Live Waste Classification",
    description="Show waste to your webcam and the model will predict its type in real-time."
)

interface.launch()