Quevedo Logo
Quevedo Logo

Quevedo

Quevedo is a python tool for creating, annotating and managing datasets of graphical languages, with a focus on the training and evaluation of machine learning algorithms for their recognition.

Quevedo is part of the VisSE project. The code can be found at GitHub, and detailed documentation here.

Features

Installation

Quevedo requires python >= 3.7, and can be installed from PyPI:

$ pip install quevedo

Or, if you want any extras, like the web interface:

$ pip install quevedo[web]

Or directly from the wheel in the release file:

$ pip install quevedo-{version}-py3-none-any.whl[web]

You can test that quevedo is working

To use the neural network module, you will also need to install darknet.

Usage

To create a dataset:

$ quevedo -D path/to/new/dataset create

Then you can cd into the dataset directory so that the -D option is not needed.

You can also download an example dataset from this repository (examples/toy_arithmetic), or peruse our Corpus of Spanish Signwriting.

To see information about a dataset:

$ quevedo info

To launch the web interface (you must have installed the “web” extra):

$ quevedo web

For more information, and the list of commands, run quevedo --help or quevedo <command> --help or see here.

Development

To develop on quevedo, we use poetry as our environment, dependency and build management tool. In the quevedo code directory, run:

$ poetry install

Then you can run quevedo with

$ poetry run quevedo

Dependencies

Quevedo makes use of the following open source projects:

Additionally, we use the toml and forcelayout libraries, and build our documentation with mkdocs.

About

Quevedo is licensed under the Open Software License version 3.0.

The web interface includes a copy of preactjs for ease of offline use, distributed under the MIT License.

Quevedo is part of the project “Visualizando la SignoEscritura” (Proyecto VisSE, Facultad de Informática, Universidad Complutense de Madrid) as part of the program for funding of research projects on Accesible Technologies financed by INDRA and Fundación Universia. An expert system developed using Quevedo is described in this article.

VisSE team