import gradio as gr
import cv2
import numpy as np
from tensorflow.keras.models import load_model
from tensorflow.keras.applications.mobilenet_v2 import preprocess_input

# Load the pre-trained model
model = load_model("C:\Users\User\Downloads\klasifikasi_jerawat_model_klmpk2.keras")

#D:\SEMS 3\AI\klasifikasi_jerawat_model_klmpk2.keras

# Define acne labels
acne_labels = {
    0: 'papules',
    1: 'nodules',
    2: 'pustules',
    3: 'comedones'
}

# Define the detection function
def detect_acne(image, threshold=0.5):
    # Resize and preprocess the image
    image_resized = cv2.resize(image, (224, 224))
    input_data = preprocess_input(np.expand_dims(image_resized, axis=0))

    # Model prediction
    predictions = model.predict(input_data)

    # Process predictions
    detections = []
    for i, prediction in enumerate(predictions[0]):
        if prediction > threshold:
            detections.append({
                'class': acne_labels[i],
                'confidence': float(prediction)
            })

    # Annotate image if detections exist
    annotated_image = image.copy()
    if detections:
        for detection in detections:
            # Simulate bounding box coordinates
            height, width, _ = image.shape
            xmin, ymin = int(width * 0.25), int(height * 0.25)
            xmax, ymax = int(width * 0.75), int(height * 0.75)

            # Draw bounding box
            color = (203, 0, 203)  # Purple-red
            cv2.rectangle(annotated_image, (xmin, ymin), (xmax, ymax), color, 2)

        return annotated_image, f"Detected acne: {detections}"
    else:
        return annotated_image, "No acne detected. Congrats!"

# Define Gradio interface
interface = gr.Interface(
    fn=detect_acne,
    inputs=gr.Image(type="numpy", label="Upload an image"),  # Input as image
    outputs=[
        gr.Image(type="numpy", label="Annotated Image"),     # Output as image
        gr.Textbox(label="Detection Result")                # Output as text
    ]
)

# Launch the app
interface.launch()