app.py

import gradio as gr
from gradio_client import Client, file
import numpy as np
from PIL import Image

def process_image(image):
    client = Client("prs-eth/marigold")

    # Convert PIL image to file-like object
    image.save("/tmp/temp_image.jpg")
    with open("/tmp/temp_image.jpg", "rb") as img_file:
        img = file(img_file)

    # Setup parameters
    ensemble_size = 10
    denoising_steps = 15
    processing_res = "0"
    match_input_res = True
    color_map = None  # Assuming no color map is desired

    # Perform the prediction
    result = client.predict(
        img,
        ensemble_size,
        denoising_steps,
        processing_res,
        match_input_res,
        color_map,
        api_name="/submit_depth_fn"
    )

    # Assuming 'depth_np' is returned as a numpy array
    if result and 'depth_np' in result:
        # Convert numpy array back to image for display
        depth_array = np.array(result['depth_np'])
        depth_image = Image.fromarray((depth_array * 255).astype(np.uint8))
        return depth_image
    else:
        raise ValueError("No valid output received or error in processing")

# Create the Gradio interface
iface = gr.Interface(
    fn=process_image,
    inputs=gr.inputs.Image(),
    outputs=gr.outputs.Image(),
    title="Depth Map Estimation",
    description="Upload an image to estimate the depth map using Marigold API."
)

# Run or deploy the interface
iface.launch()