Installation¶

tfaip is available on pypi and can thus be installed with pip:

pip install tfaip

Prerequisites¶

You need to install the following on your system:

Python 3.7 or later, including
the python development packages (on Ubuntu apt install python3.7 python3.7-dev, if not available in your distribution, use the deadsnakes repo for ubuntu based distros)
a gcc compiler and all other build essentials (on Ubuntu apt install build-essential)
(optional) cuda/cudnn libs for GPU support, see tensorflow for the versions which are required/compatible.

Setup in a Virtual Environment¶

Setup your venv and install:

<python> must be replaced with python3.7 or later (check <python> --version before):

virtualenv -p <python> PATH_TO_VENV
source PATH_TO_VENV/bin/activate
pip install -U pip  # recommended to get the latest version of pip
pip install tfaip

Possible bugs¶

pycocotools¶

Currently as of numpy 1.20, pycocotools are falsely compiled, run

pip uninstall -y pycocotools
pip install --no-cache-dir pycocotools

to fix your venv by reinstalling and recompiling the pycocotools.

Tests¶

tfaip provides a set of tests which are listed in the test directory using the unittest framework.

Run the tests¶

Call python -m unittest or pytest to run the tests from the command line.

In PyCharm: Right-click on test select Run 'Unittests in test', or “Shift-Shift”, write “run” and choose Run 'Unittests in test'

CI/CD¶

All tests will automatically run for CI/CD.

Development Setup¶

For development support, clone the code, install the requirements in a fresh virtual environment, and link the tfaip source to the virtualenv:

git clone https://github.com/Planet-AI-GmbH/tfaip.git  # or git clone git@github.com:Planet-AI-GmbH/tfaip.git
cd tfaip
virtualenv -p python3 venv
source venv/bin/activate
pip install -U pip  # recommended to get the latest version of pip
pip install -r requirements.txt
python setup.py develop