app.py

import os
import socket
import time
import gradio as gr
import numpy as np
from PIL import Image
from loguru import logger
import cv2
import torch

# Import MMRotate
from mmrotate.models import build_detector
from mmdet.apis import init_detector, inference_detector
from mmrotate.apis import inference_detector_by_patches
from mmrotate.datasets import DOTADataset

# Default size for model
IMG_SIZE = 1024
OVERLAP = 192
MARGIN = OVERLAP / 2

# depends on the GPU memory
BATCH_SIZE = 16

# CLASSES
CLASSES = ['ship',]

# Choose to use a config and initialize the detector
config_file = 'redet_re50_refpn_1x_dota_ms_rr_le90.py'

# Setup a checkpoint file to load
weights_file = 'weights/best_mAP_epoch_20.pth'

# check if GPU if available
device = torch.device("cuda:0" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu")
logger.info(f"Using device: {device}")

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

# build the model from a config file and a checkpoint file
model = init_detector(config_file, weights_file, device)

# 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.")

    start_time = time.time()
    if img.shape[0] > IMG_SIZE or img.shape[1] > IMG_SIZE:
        #print("Running inference_detector_by_patches")
        result = inference_detector_by_patches(model, img, sizes=[IMG_SIZE], steps=[IMG_SIZE - 2 * MARGIN], ratios=[1.0], merge_iou_thr=0.3, bs=BATCH_SIZE)
    else:
        #print("Running inference_detector")
        result = inference_detector(model, img)
    end_time = time.time()
    #print(result)

    # total number of predictions
    infos = np.sum(result[0][:, -1] > threshold)

    img_preds = model.show_result(img, result, score_thr=threshold, show=False)
    return img_preds, img.shape, infos, end_time - start_time


# Define example images and their true labels for users to choose from
example_data = [
    ["./demo/82f13510a.jpg", 0.75],
    ["./demo/836f35381.jpg", 0.75],
    ["./demo/848d2afef.jpg", 0.75],
    ["./demo/Satellite_Image_Marina_New_Zealand.jpg", 0.4],
    ["./demo/Pleiades_HD15_Miami_Marina.jpg", 0.4],
    # Add more example images and labels as needed
]

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

TITLE = "Ship Detection on Optical Satellite image"

# Define the Gradio Interface
demo = gr.Blocks(title=TITLE, css=css).queue()
with demo:
    gr.Markdown(f"<h3><center>{TITLE}<center><h3>")

    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 ships", 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 Airbus Ship Detection available on Kaggle. The associated license is <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>")


if os.path.exists('/.dockerenv'):
    print('Running inside a Docker container')

    # Launch the interface on MacOS
    hostname = socket.gethostname()

    demo.launch(
        server_name=hostname, 
        inline=False, 
        server_port=7860, 
        debug=True
    )
else:
    print('Not running inside a Docker container')
    demo.launch(
        inline=False, 
        server_port=7860, 
        debug=False
    )