import gradio as gr
from PIL import Image
import torch
import torchvision.transforms as transforms
from model import RetinaNet  # Import your RetinaNet model definition

# Define the image transformation pipeline
image_transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

# Load the model
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = RetinaNet(num_classes=2).to(device)
model.load_state_dict(torch.load("retinanet_best_model.pth", map_location=device))
model.eval()

# Prediction function
def predict_image(image):
    # Preprocess the image
    img = Image.fromarray(image).convert('RGB')  # Convert Gradio input to PIL Image
    input_tensor = image_transform(img).unsqueeze(0).to(device)

    # Perform inference
    with torch.no_grad():
        prediction = model(input_tensor.float())
        sum_value = abs(torch.sum(prediction[0]))
        p_true = abs(prediction[0][0])
        p_false = abs(prediction[0][1])

    # Interpret the prediction
    if p_true > 0.7:
        result = "Accepted"
        confidence = float(p_true)
    else:
        result = "Rejected"
        confidence = float(p_false)

    return f"Result: {result}, Confidence: {confidence:.2f}"

# Create the Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# RetinaNet Model Prediction")
    with gr.Row():
        image_input = gr.Image(label="Upload Image", type="numpy")
        output_text = gr.Textbox(label="Prediction Result")
    predict_button = gr.Button("Predict")
    predict_button.click(predict_image, inputs=image_input, outputs=output_text)

# Launch the app
demo.launch()