Encoder-Decoder Caption Service

This service is responsible for image captioning using an encoder-decoder network on image streams and the output is a vekg of each image.

Commands Stream

Inputs

...

Outputs

...

Data Stream

Expect an data event with at least the following fields, as the following example:

{
    "id": "abc-123abc-123abc-123abc-123abc-123abc-123",
    "image_url": "c8d025d3-8c3a-460c-a6f5-cabb7b179807",
    "data_flow": [
        ["object-detection-data"],
        ["color-detection-data"],
        ["gb-data"]
    ],
    "data_path": [],
    "vekg": {},
}

Enrich the event data with the following fields, as the following example:

 "vekg": {
    "nodes": [
        [
            "02fad514-6c93-4b09-969b-666e9cd52799",
             {
                 "id": "02fad514-6c93-4b09-969b-666e9cd52799",
                 "caption": "car moving down the street",
              }
        ],
    ]
  },
  "data_path": ["object-detection-data",],
}

Installation

Configure .env

Copy the example.env file to .env, and inside it replace SIT_PYPI_USER and SIT_PYPI_PASS with the correct information.

Installing Dependencies

Using pipenv

Run $ pipenv shell to create a python virtualenv and load the .env into the environment variables in the shell.

Then run: $ pipenv install to install all packages, or $ pipenv install -d to also install the packages that help during development, eg: ipython. This runs the installation using pip under the hood, but also handle the cross dependency issues between packages and checks the packages MD5s for security mesure.

Using pip

To install using pip directly, one needs to use the --extra-index-url when running the pip install command, in order for to be able to use our private Pypi repository.

Load the environment variables from .env file using source load_env.sh.

To install from the requirements.txt file, run the following command:

$ pip install --extra-index-url https://${SIT_PYPI_USER}:${SIT_PYPI_PASS}@sit-pypi.herokuapp.com/simple -r requirements.txt

Running

Enter project python environment (virtualenv or conda environment)

ps: It's required to have the .env variables loaded into the shell so that the project can run properly. An easy way of doing this is using pipenv shell to start the python environment with the .env file loaded or using the source load_env.sh command inside your preferable python environment (eg: conda).

Then, run the service with:

$ ./object_detection_service/run.py

Testing

Run the script run_tests.sh, it will run all tests defined in the tests directory.

Also, there's a python script at ./object_detection_service/send_msgs_test.py to do some simple manual testing, by sending msgs to the service stream key.

Docker

Build

Build the docker image using: docker-compose build

ps: It's required to have the .env variables loaded into the shell so that the container can build properly. An easy way of doing this is using pipenv shell to start the python environment with the .env file loaded or using the source load_env.sh command inside your preferable python environment (eg: conda).

Run

Use docker-compose run --rm service to run the docker image

Gitlab CI auto-build and tests

This is automatically enabled for this project (using the .gitlab-ci.yml present in this project root folder).

By default it will build the Dockerfile with every commit sent to the origin repository and tag it as 'dev'.

Afterwards, it will use this newly builty image to run the tests using the ./run_tests.sh script.

But in order to make the automatic docker image build work, you'll need to set the SIT_PYPI_USER and SIT_PYPI_PASS variables in the Gitlab CI setting page: Object Detection Service CI Setting Page.

And, in order to make the automatic tests work, you should also set the rest of the environement variables required by your service, in the this projects .gitlab-ci.yml file, in the variables section. But don't add sensitive information to this file, such as passwords, this should be set through the Gitlab CI settings page, just like the SIT_PYPI_USER.

Benchmark Tests

To run the benchmark tests one needs to manually start the Benchmark stage in the CI pipeline, it shoud be enabled after the tests stage is done. Only by passing the benchmark tests shoud the image be tagged with 'latest', to show that it is a stable docker image.