app.py

import gradio as gr
from ultralytics import YOLO
import cv2
import numpy as np
import os
import requests
import torch
import datetime
import subprocess
import spaces  # Ensure this import is correct and the module is available

CUSTOM_CSS = """
#output_box textarea {
    font-family: IBM Plex Mono, ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, "Liberation Mono", "Courier New", monospace;
}
"""

# Ensure the model file is in the correct location
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).to(device)

def process_image(image):
    # Convert image to the format YOLO model expects
    image = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
    results = docseg_model(image)

    # Extract annotated image from results
    annotated_img = results[0].plot()
    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}: {box.conf:.2f}" for box in results[0].boxes]
    )

    return annotated_img, detected_areas_labels

zero = torch.Tensor([0]).cuda()
print(zero.device) # <-- 'cuda:0' if GPU is available, otherwise 'cpu'

@spaces.GPU
def run_gpu() -> str:
    print(zero.device) # <-- 'cuda:0' 🤗
    output: str = ""
    try:
        output = subprocess.check_output(["nvidia-smi"], text=True)
    except FileNotFoundError:
        output = "nvidia-smi failed"
    comment = (
        datetime.datetime.now().replace(microsecond=0).isoformat().replace("T", " ")
    )
    return f"# {comment}\n\n{output}"

def run(check: bool) -> str:
    if check:
        return run_gpu()
    else:
        comment = (
            datetime.datetime.now().replace(microsecond=0).isoformat().replace("T", " ")
        )
        return f"# {comment}\n\nThis is running on CPU\n\nClick on 'Run on GPU' below to move to GPU instantly and run nvidia-smi"

output = gr.Textbox(
    label="Command Output", max_lines=32, elem_id="output_box", value=run(False)
)

with gr.Blocks(css=CUSTOM_CSS) as demo:
    gr.Markdown("#### `zero-gpu`: how to run on serverless GPU for free on Spaces 🔥")

    output.render()

    check = gr.Checkbox(label="Run on GPU")

    check.change(run, inputs=[check], outputs=output, every=1)

# 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]
    )

demo.queue().launch(show_api=False)
interface.launch()

if __name__ == "__main__":
    demo.launch()
    interface.launch()