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.

Installation

We recommend to install Wentral into a virtualenv. After the virtualenv is activated execute python setup.py install. Dependencies will be installed automatically.

Web service

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.

Benchmarks

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.

Development

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.

Testing

Python

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

JavaScript

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.

CI

The CI setup in the GitLab repository runs both Python tests and JavaScript linting. It's configured via .gitlab-ci.yml.

License

Wentral is Free and Open Source software distributed under the terms of MIT license (see LICENSE.txt for more details).