app.py

from gradio_client import Client
from PIL import Image
import gradio as gr
import tempfile
import os


# Get Hugging Face Token from environment variables
# This token is crucial for accessing private spaces
HF_TOKEN = os.getenv("HF_TOKEN")

# 1️⃣ Connect to the private Space
# Ensure the URL is correct and the HF_TOKEN has permissions to access this private space.
client = Client("Insoore/damage_detector_insoore_gpu", hf_token=HF_TOKEN)

print("\n*** Private‑Space API ***")
# Print the API definition of the private space for debugging purposes
print(client.view_api())
print("*** end API print ***\n")

def call_private_api(pil_img: Image.Image):
    """
    Calls the private damage detection API with the uploaded image and parameters.

    Args:
        pil_img (Image.Image): The input image uploaded by the user.

    Returns:
        Image.Image: The processed image with detected damages from the private API.
    """
    # Save the user-uploaded image to a temporary JPEG file.
    # This is necessary because gradio_client expects a file path for image inputs.
    file_path = None # Initialize file_path
    try:
        with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp:
            pil_img.convert("RGB").save(tmp, "JPEG", quality=95)
            file_path = tmp.name  # Store the path to the temporary file

        # DEBUG: Confirming type and value of file_path
        print(f"DEBUG: Type of file_path: {type(file_path)}")
        print(f"DEBUG: Value of file_path: {file_path}")

        # Pass the file_path string directly as the image input.
        # Temporarily removed all optional parameters to isolate the image input issue.
        result_path = client.predict(
            image=file_path, # Pass the file path directly as a string
            api_name="/predict",
        )

        # result_path is a downloaded file path from the private API's output.
        # Open this file and return it as a PIL Image.
        return Image.open(result_path)
    except Exception as e:
        # Basic error handling: print the error and re-raise or return a placeholder
        print(f"Error calling private API: {e}")
        # In a real application, you might want to return a blank image or an error message
        raise gr.Error(f"Failed to detect damages: {e}")
    finally:
        # Clean up the temporary file after the prediction
        if file_path and os.path.exists(file_path):
            os.remove(file_path)


            

# 3️⃣ Set up the Gradio interface for the public space
with gr.Blocks() as demo:
    gr.Markdown("# Damage Detector")
    gr.Markdown("Upload an image to detect damages using a private AI model.")
    
    with gr.Row():
        inp = gr.Image(type="pil", label="Upload an image of a vehicle")
        out = gr.Image(type="pil", label="Detected damages highlighted")
    
    # Button to trigger the API call
    gr.Button("Detect Damages").click(
        fn=call_private_api,
        inputs=inp,
        outputs=out,
        api_name="detect_damages_public" # Optional: give a public API name for this function
    )

# Launch the Gradio demo
# show_error=True will display exceptions in the UI
# share=True is for creating a public link (though Hugging Face Spaces handle this automatically)
demo.launch(show_error=True, share=True)