app.py

import os
import socket
import time
import gradio as gr
import numpy as np
from PIL import Image, ImageDraw
import base64
import requests
import json

# API for inferences
DL4EO_API_URL = "https://dl4eo--object-predict-raster.modal.run"

# Auth Token to access API
DL4EO_API_KEY = os.environ['DL4EO_API_KEY']

# width of the boxes on image
LINE_WIDTH = 2

# Check Gradio version
print(f"Gradio version: {gr.__version__}")

# Define the inference function
def predict_image(img, threshold):
    
    if isinstance(img, Image.Image):
        img = np.array(img)
    
    if not isinstance(img, np.ndarray) or len(img.shape) != 3 or img.shape[2] != 3:
        raise BaseException("predit_image(): input 'img' shoud be single RGB image in PIL or Numpy array format.")

    # Encode the image data as base64
    image_base64 = base64.b64encode(np.ascontiguousarray(img)).decode()
    
    # Create a dictionary representing the JSON payload
    payload = {
        'image': image_base64,
        'shape': img.shape,
        'threshold': threshold,
    }

    headers = {
        'Authorization': 'Bearer ' + DL4EO_API_KEY,
        'Content-Type': 'application/json'  # Adjust the content type as needed
    }

    # Send the POST request to the API endpoint with the image file as binary payload
    response = requests.post(DL4EO_API_URL, json=payload, headers=headers)
    
    # Check the response status
    if response.status_code != 200:
        raise Exception(
            f"Received status code={response.status_code} in inference API"
        )
            
    json_data = json.loads(response.content)
    image_data = json_data['image']
    shape = json_data['shape']
    infos = json_data['infos']
    duration = json_data['duration']

    img_data_in = base64.b64decode(json_data['image']) 
    np_img = np.frombuffer(img_data_in, dtype=np.uint8).reshape(shape)
    pil_img = Image.fromarray(np_img)

    return pil_img, shape, infos, duration


# Define example images and their true labels for users to choose from
example_data = [
    ["./demo/demo.jpg", 0.8],
    ["./demo/dota_demo.jpg", 0.8],
    ["./demo/Satellite_Image_Marina_New_Zealand.jpg", 0.8],
    ["./demo/Pleiades_HD15_Miami_Marina.jpg", 0.8],
    ["./demo/SPOT_Storage.jpg", 0.8],
    ["./demo/airport01.jpg", 0.8],
    ["./demo/Pleiades_Neo_Tucson_USA.jpg", 0.8],
    # Add more example images and labels as needed
]

# Define CSS for some elements
css = """
  .image-preview {
    height: 820px !important; 
    width: 800px !important;
  } 
"""

TITLE = "Oriented bounding boxes detection on Optical Satellite images"

# Define the Gradio Interface
demo = gr.Blocks(title=TITLE, css=css).queue()
with demo:
    gr.Markdown(f"<h1><center>{TITLE}<center><h1>")

    with gr.Row():
        with gr.Column(scale=0):
            input_image = gr.Image(type="pil", interactive=True)
            run_button = gr.Button(value="Run")
            with gr.Accordion("Advanced options", open=True):
                threshold = gr.Slider(label="Confidence threshold", minimum=0.0, maximum=1.0, value=0.25, step=0.01)
                dimensions = gr.Textbox(label="Image size", interactive=False)
                detections = gr.Textbox(label="Predicted objects", interactive=False)
                stopwatch = gr.Number(label="Execution time (sec.)", interactive=False, precision=3)

        with gr.Column(scale=2):
            output_image = gr.Image(type="pil", elem_classes='image-preview', interactive=False)

    run_button.click(fn=predict_image, inputs=[input_image, threshold], outputs=[output_image, dimensions, detections, stopwatch])
    gr.Examples(
        examples=example_data,
        inputs = [input_image, threshold],
        outputs = [output_image, dimensions, detections, stopwatch],
        fn=predict_image,
        cache_examples=True,
        label='Try these images!'
    )

    gr.Markdown("""
                <p>This demo is provided by <a href='https://www.linkedin.com/in/faudi/'>Jeff Faudi</a> and <a href='https://www.dl4eo.com/'>DL4EO</a>. 
                This model is based on the <a href='https://github.com/open-mmlab/mmrotate'>MMRotate framework</a> which provides oriented bounding boxes. 
                We believe that oriented bouding boxes are better suited for detection in satellite images. This model has been trained on the 
                <a href='https://captain-whu.github.io/DOTA/dataset.html'>DOTA dataset</a> which contains 15 classes: plane, ship, storage tank, 
                baseball diamond, tennis court, basketball court, ground track field, harbor, bridge, large vehicle, small vehicle, helicopter, 
                roundabout, soccer ball field and swimming pool. </p><p>The associated licenses are 
                <a href='https://about.google/brand-resource-center/products-and-services/geo-guidelines/#google-earth-web-and-apps'>GoogleEarth fair use</a> 
                and <a href='https://creativecommons.org/licenses/by-nc-sa/4.0/deed.en'>CC-BY-SA-NC</a>. This demonstration CANNOT be used for commercial puposes. 
                Please contact <a href='mailto:jeff@dl4eo.com'>me</a> for more information on how you could get access to a commercial grade model or API. </p>
                """)

demo.launch(
    inline=False, 
    show_api=False,
    debug=False
)