app.py

import gradio as gr
from ultralytics import YOLO
import cv2
import numpy as np
import os
import requests
import torch

# Load the model file
model_path = "yolov8x-doclaynet-epoch64-imgsz640-initiallr1e-4-finallr1e-5.pt"
if not os.path.exists(model_path):
    # Download the model file if it doesn't exist
    model_url = "https://huggingface.co/DILHTWD/documentlayoutsegmentation_YOLOv8_ondoclaynet/resolve/main/yolov8x-doclaynet-epoch64-imgsz640-initiallr1e-4-finallr1e-5.pt"
    response = requests.get(model_url)
    with open(model_path, "wb") as f:
        f.write(response.content)

# Load the document segmentation model
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
docseg_model = YOLO(model_path)  # Remove .to(device) to let ultralytics auto-detect

def process_image(image):
    try:
        # Convert image to the format YOLO model expects
        image = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
        results = docseg_model.predict(image)  # Use predict for inference
        result = results[0]  # Get the first (and usually only) result
        
        # Extract annotated image from results
        annotated_img = result.plot()  # Simplified plotting
        annotated_img = cv2.cvtColor(annotated_img, cv2.COLOR_BGR2RGB)

        # Prepare detected areas and labels as text output
        detected_areas_labels = "\n".join(
            [f"{box.label.upper()}: {box.conf:.2f}" for box in result.boxes]  # Uppercase labels
        )
    except Exception as e:
        return None, f"Error during processing: {e}"  # Error handling

    return annotated_img, detected_areas_labels

# Define the Gradio interface
with gr.Blocks() as interface:
    gr.Markdown("### Document Segmentation using YOLOv8")
    input_image = gr.Image(type="pil", label="Input Image")
    output_image = gr.Image(type="pil", label="Annotated Image")
    output_text = gr.Textbox(label="Detected Areas and Labels")

    gr.Button("Run").click(
        fn=process_image,
        inputs=input_image,
        outputs=[output_image, output_text]
    )

# Launch the interface (remove the conditional launch)
interface.launch(share=True)  # Allow sharing for easier debugging