A frontend for machine learning models that detect objects in web pages, that can be used to:
- Measure and visualize the performance of object detectors on datasets,
- Expose an object detector as an HTTP web service.
We recommend to install Wentral into a virtualenv. After the virtualenv is
activated execute python setup.py install. Dependencies will be installed
automatically.
To make a web service from a detector.Class loading the weights from
weights/file run:
$ wentral ws -d detector.Class -w weights/file
See CLI docs for more info on usage.
To benchmark detector.Class on a data/set run:
$ wentral bm [-v] -d detector.Class -w weights/file data/set
See CLI docs for more info on usage.
Most common scenario will be implementing detectors to use with Wentral. The API docs has more detail on this.
You are also welcome to contribute to Wentral itself. Make sure the tests still pass and the coverage is not reduced. Make sure to follow eyeo coding style to make reviews simpler.
We use Tox for testing Python code and Python linting. Install Tox with pip if you haven't already and then run the tests with:
$ tox
There's also a small amount of JavaScript in this repo. Unfortunately it has no tests, but you can lint it using ESLint (more info on eyeo eslint config here). Make sure you have ESLint and eyeo config installed:
$ npm install -g eslint eslint-config-eyeo
and then run:
$ eslint wentral/vis_ui/visualization.js
You only need to do it if you changed that file.
The CI setup in the GitLab repository runs both Python tests and JavaScript linting. It's configured via .gitlab-ci.yml.
Wentral is Free and Open Source software distributed under the terms of MIT license (see LICENSE.txt for more details).