DeployAI Object Detection

This repository contains a deep learning model for detecting objects from very high resolution optical remote sensing images using YOLOv8 Oriented Bounding Box, which is served using a web interface using Flask.

Key features

The model allows detecting the following objects

{
    "plane": 0,
    "ship": 1,
    "storage-tank": 2,
    "baseball-diamond": 3,
    "tennis-court": 4,
    "basketball-court": 5,
    "ground-track-field": 6,
    "harbor": 7,
    "bridge": 8,
    "large-vehicle": 9,
    "small-vehicle": 10,
    "helicopter": 11,
    "roundabout": 12,
    "soccer-ball-field": 13,
    "swimming-pool": 14,
    "container-crane": 15,
    "airport": 16,
    "helipad": 17,
}

1. Navigate to the satellite-od-model web UI and upload the aerial or high-resolution satellite images.
2. Once the processing is completed, you can find the bounding boxes overlayed on the input images with classes of objects detected.
3. Results and intermittent files are stored in the shared folder.

Input Format

The input image should be of size $1024\times1024\times3$, which means RGB images. The image extension can be tif. Via the webapp, the user can upload multiple images, where objects will be detected by the model. After prediction termination, the user can download the results from the webapp as a compressed file.

Output Format

The output is a text file that contains one or multiple lines. Each line is associated with one detected object. A line starts with the class ID, the coordinates of the bounding box in as floats and the final float number is the confidence of the model for this detected object. The following is an example of an output file

10 0.769302 0.0571918 0.771534 0.0672572 0.792736 0.0625548 0.790504 0.0524893 0.76106
10 0.616684 0.486967 0.617148 0.478474 0.597924 0.477423 0.59746 0.485916 0.750721
10 0.510295 0.481979 0.510747 0.471428 0.490472 0.470558 0.490019 0.481109 0.548859
10 0.522158 0.0857164 0.524665 0.095236 0.546002 0.0896183 0.543495 0.0800987 0.545464
10 0.132021 0.358989 0.143523 0.356336 0.138069 0.332698 0.126568 0.335351 0.537943
9 0.570139 0.719797 0.570226 0.732347 0.604308 0.732111 0.604221 0.719561 0.327334
10 0.930733 0.603296 0.941956 0.603665 0.94258 0.584695 0.931358 0.584326 0.317335
10 0.283544 0.659585 0.294072 0.659736 0.294374 0.638583 0.283846 0.638433 0.301854
10 0.792477 0.0173265 0.800817 0.0147654 0.79538 -0.00294156 0.78704 -0.000380462 0.300901
10 0.808204 0.0153187 0.815899 0.0122031 0.809846 -0.00274596 0.802151 0.000369608 0.294774
10 0.890847 0.281562 0.891716 0.273191 0.875498 0.271507 0.874628 0.279878 0.283491
10 0.989942 0.247576 0.999541 0.251096 1.00361 0.240008 0.994008 0.236487 0.27634

Important

If no objects are detected for an image, no output file is produced.

Repository Content

.
├── .github/
│ └── workflows/
│     └── build-test-image.yaml
├── README.md
├── Dockerfile
├── model.proto (Ptorobuf v3 specs)
├── requirements.txt
├── serve.py (gRPC service)
├── .gitignore
├── .dockerignore
├── .pylintrc
└── app/ (DL model)
    ├── init.py
    ├── config.py
    ├── utils.py
    ├── inference.py
    └── weights/
        └── best.pt

Local Development

• In a terminal, clone the repository

git clone https://github.com/AlbughdadiM/depai-yolov8-obb.git

• Go to the repository directory

cd depai-yolov8-obb

• If the files model_pb2_grpc.py and model_pb2.py are not there, generate them using

python3.10 -m grpc_tools.protoc -I. --python_out=. --grpc_python_out=. model.proto

• Build the docker image

docker build . -t object-detection:v0.1

• Create a container from the built image

docker run --name=test -v ./test-data:/data --env SHARED_FOLDER_PATH=/data -p 8061:8061 -p 8062:8062 object-detection:v0.1

• Run the pytest

pytest test_image_processor.py

Container Registry

• Generate a personal access token: Github account settings > Developer settings > Personal access tokens (classic). Generate a token with the read:package scope.

• In a terminal, login to container registry using

docker login ghcr.io -u USERNAME -p PAT

• Pull the image

docker pull ghcr.io/albughdadim/depai-yolov8-obb:v0.1

• Create a container

docker run --name=test -v ./test-data:/data --env SHARED_FOLDER_PATH=/data -p 8061:8061 -p 8062:8062 ghcr.io/albughdadim/depai-yolov8-obb:v0.1

How TO Use The Model Without GUI

import grpc
import model_pb2
import model_pb2_grpc


def run():
    channel = grpc.insecure_channel("localhost:8061")
    stub = model_pb2_grpc.ImageProcessorStub(channel)
    # As we can see, the model accepts multiple images in a list
    response = stub.ProcessImage(
        model_pb2.ImageRequest(  # pylint: disable=E1101
            input_image_paths=[
                "/data/patch_250.tif",
                "/data/patch_420.tif",
            ]
        )
    )
    print("Output Result Path: " + response.entries[0].result_path)
    print("Output Result Path: " + response.entries[1].result_path)


if __name__ == "__main__":
    run()

This is a list that has the same length of the input one. In each element of the list, there is a dictinary with three keys:

• image_path: Input image path.
• processed: Boolean to indicate if an image is procssed or not.
• result_path: Output txt file containing bounding boxes of detected objects.