Installation and Usage#

Why using HALF ?#

Active learning methods helps to reduce labelling cost and time by selecting the most interesting points to label in the data according to what your model has learned up until now.

HALF is agnostic of the model and the dataset, also allowing for a full customisation of the full active learning pipeline, including the training of the model.

Installation#

HALF is dockerized, meaning you only need to have Docker installed and our repository in local to be able to use it.

  1. First, you need to clone the repository :

git clone

See also

(Optional) Check that the version of the pytorch image used in the dockerfile found in the folder Framework matches your nvidia drivers.

  1. Add your custom Dataset.py and Model.py to import your dataset and models in the Extensions folder. You can find examples in the currently existing files.

  2. Customise the config.yaml in the Configs folder according to your needs.

  3. Launch the docker with the following instructions from the root` folder :

docker compose up

  1. Admire the process of active learning launching !

Usage#

Once the docker attached, you can simply launch the process with :

python Launcher/main.py