Segmantation and recognition service (segandrec-service).

Description

ITMO project for microservices. Provides the backend service for bottles segmentation and (not yet) mark detection.

Models used:

• Mask-RCNN for segmentation
• Exception (probably) for mark detection

Ease-Of-Use

First of all, let's download the checkpoints of pre-trained mask-rcnn and unpack them into ./MaskRCNN/weights folder. You must make sure that files ./MaskRCNN/weights/variables/ and ./MaskRCNN/weights/saved_model.pb are available from here by exact same paths.

Tensorflow checkpoints are publicly available by link: https://drive.google.com/file/d/1Y2YSaUfsKISgbVzDlibVWYbj4bTfUN_1/view?usp=sharing

Next big thing is changing the host file to contain the address you will need (e.g. http://localhost:5000)

Run server locally:

Let's execute the following command in order to install all necessarry python libraries:

pip3 install -r requirements.txt

Now, run the flask backend program by executing:

python3 app.py or flask run -h 0.0.0.0 -p 5000

Run server using docker container:

In order to build the docker image, run:

docker build -t flask .

Create a container and traverse the 5000 port onto the local machine:

docker run -d -p 5000:5000 flask

Run client:

And finally, let's execute the example client code to test if everything works ok:

python3 request_example.py

As a result, I presume you will see the picture of a bottle with whitened background on your screen.

Note: if you are running NOT from docker image, make sure that you have at least following packages installed on your system: Pillow, numpy, opencv-python, requests, tensorflow

API Documentation:

You can see more about the service API by running it on your local machine and following the swagger link: http://IP:PORT/api/v1.0/swagger_ui

Segmentation has to be performed by the following scenario: http://IP:PORT/api/v1.0/segmentation

As s DATA parameter you should be sending JSON structure containing link to image

JSON Request structure:

JSON: {
    image: "link"
}

Response structure for successful segmentations:

JSON: {
    status: 200
    segmentation: "link"
    mask: "link"
}

Response structure for errors 400, 404, 405 and 500:

JSON: {
    status: int
    error: 'Text of error'
}