NaoTH-DeepLearning

This repo contains tools for datasets, learning, compiling and evaluating neural networks. This is for all deep learning done as part of the RoboCup SPL efforts of the Berlin United Team.

Each folder corresponds to one project. A project might be whistle detection, ball detection on patches, yolo ball detection, object detection on full images, etc. Commonly used functions should go into the tools folder.

Installation

We prepared different requirement files if you want to work with the tensorflow projects or the pytorch projects for example ultralytics. Choose the corresponding requirements file during installation. We expect python 3.12.

python3 -m venv venv_tf
source venv_tf/bin/activate
python -m pip install -r requirements-tensorflow.txt

or

python3 -m venv venv_pytorch
source venv_pytorch/bin/activate
python -m pip install -r requirements-pytorch.txt

How to setup direnv

It's recommended to set up direnv for each subproject.

sudo apt install direnv

add this to your .bashrc if you use other shells see https://direnv.net/docs/hook.html

eval "$(direnv hook bash)"

create a .envrc file in a subproject folder. For example like this for a project that uses ultralytics

export VAT_API_URL=http://localhost:8000/
export VAT_API_TOKEN=<your-token-here>

source ../venv_pytorch/bin/activate
unset PS1

after every change of .envrc you need to run

direnv allow