app.py

import os
import socket
import gradio as gr
import numpy as np
from PIL import Image, ImageDraw
from pathlib import Path
from loguru import logger
import cv2
import torch
import time
import base64
import requests
import json
 
# API for inferences
DL4EO_API_URL = "https://dl4eo--groundingdino-predict.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 modules version
logger.info(f"Gradio version: {gr.__version__}")

# Define the inference function
def predict_image(image, text_prompt, box_threshold, text_threshold):
    
    # Resize the image to the new size
    #image = image.resize((image.size[0] * 2, image.size[1] * 2))

    if isinstance(image, Image.Image):
        img = np.array(image)
    
    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.")

    #width, height = img.shape[0], img.shape[1]

    # 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,
        'text_prompt': text_prompt,
        'box_threshold': box_threshold,
        'text_threshold': text_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: {response.text}"
        )
            
    json_data = json.loads(response.content)
    duration = json_data['duration']
    boxes = json_data['boxes']

    # drow boxes on image
    draw = ImageDraw.Draw(image)

    for box in boxes:
        left, top, right, bottom = box

        if left <= 0: left = -LINE_WIDTH 
        if top <= 0: top = top - LINE_WIDTH 
        if right >= img.shape[0] - 1: right = img.shape[0] - 1 + LINE_WIDTH
        if bottom >= img.shape[1] - 1: bottom = img.shape[1] - 1 + LINE_WIDTH

        draw.rectangle([left, top, right, bottom], outline="red", width=LINE_WIDTH)
    
    return image, str(image.size), len(boxes), duration


# Define example images and their true labels for users to choose from
example_data = [
    ["./demo/Pleiades_Neo_Tucson_USA.jpg", 'plane', 0.24, 0.24],
    ["./demo/Pleiades_Neo_Tucson_USA.jpg", 'building', 0.24, 0.24],
    #["./demo/Pleiades_Neo_Tucson_USA.jpg", 'tree', 0.24, 0.24],
    #["./demo/two-dogs-with-a-stick.jpg", "dog", 0.25, 0.25],
    #["./demo/airport01.jpg", "aircraft", 0.25, 0.25],
    #["./demo/SPOT_Storage.jpg", "storage", 0.25, 0.25],
    #["./demo/Satellite_Image_Marina_New_Zealand.jpg", "ship", 0.25, 0.25],
    ["./demo/Pleiades_HD15_Miami_Marina.jpg", "motorboat", 0.3, 0.0],
    ["./demo/Pleiades_HD15_Miami_Marina.jpg", "palm tree", 0.15, 0.3],
    ["./demo/Pleiades_HD15_Miami_Marina.jpg", "building", 0.3, 0.0],
]

# Define CSS for some elements
css = """
  .image-preview {
    height: 820px !important; 
    width: 800px !important;
  } 
"""
TITLE = "Open 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, scale=1)
            text_prompt = gr.Textbox(label="Text prompt")
            run_button = gr.Button(value="Run", scale=0)
            with gr.Accordion("Advanced options", open=True):
                box_threshold = gr.Slider(label="Box threshold", minimum=0.0, maximum=1.0, value=0.24, step=0.01)
                text_threshold = gr.Slider(label="Text threshold", minimum=0.0, maximum=1.0, value=0.24, step=0.01)
                dimensions = gr.Textbox(label="Image size", interactive=False)
                detections = gr.Number(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, width=800, height=800)

    run_button.click(fn=predict_image, inputs=[input_image, text_prompt, box_threshold, text_threshold], outputs=[output_image, dimensions, detections, stopwatch])
    gr.Examples(
        examples=example_data,
        inputs = [input_image, text_prompt, box_threshold, text_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>. The demonstration images are Pléiades \
                images provided by CNES with distribution by Airbus DS. The model architecture and weights \
                are provided <a href='https://github.com/IDEA-Research/GroundingDINO'>Grounding DINO</a>. \
                The model has not been trained specifically on satellite imagery and should be finetuned for this task. \
                This is for demonstration only. Please contact <a href='mailto:[email protected]'>me</a> \
                for more information on how you could get access to a commercial model or API. </p>")

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