README.md

Flask-server to explore augmentations

This package contains a Flask application to render image augmentations.

Warning: It is highly insecure and is NOT recommended to run in production.

The package is organized as a proper Python package installable via pip (so that no hacks like sys.append or export PYTHONPATH are needed).
Flask app uses mymodel's augmentations as a proper python package.
No logic has been changed except that we print the transformations and their probabilities together with the image.
Many problematic places are highlighed with TODO (no parameters exposed, UX and security issues, etc.)
To improve code quality, we implemented linters + auto-formatters (black, flake, isort) and type checker (mypy).
Also, a basic test on Flask app added.
As it was asked, we have put both python packages into a docker image that is built via GitHub Actions CI/CD.
CI/CD runs on each Pull Requests to the master branch, as well as on each master commit
CI runs linter and test checks for both packages.
If CI checks passed, it then builds the image using Dockerfile are pushes it to DockerHub under the public repository artemlops/flask_augmentations.
The tag of the image is 7 digits of the commit that triggered the build. Note: the latest image does NOT exist.

Setup your current workspace: python -m venv venv && source ./venv/bin/activate && make setup
Run linter checks: make lint
Run tests: make test
Once changed the code, auto-format it: make format
To build and push docker image: make docker_build && make docker_push
To run Flask locally: make serve
To run Flask in docker: make docker_serve COMMIT=a11f7e6d (7-digit of a commit SHA).
To stop the Flask in docker: make docker_kill

Once running a Flask app, test it:

Some transforms handle black-white images incorrectly, thus throwing 500 error. Now Flask app works properly only for RGB images (see examples).
Support multi-processing: now, only 1 user can work with the app at a single moment of time.
Security: white-list domain names so that the server doesn't arbitrary requests to any host on the Internet,

Better test coverage (integration tests, etc).
Automatic deployment, if needed.
Better UI (for example, allow the user to insert url into a text frame).
Better flexibility: expose more start-up parameters (host, port, n for maximum number of transformations, etc).