Please follow each of the steps below before arriving at RVSS:
- Clone the repository
- Download the starter data
- Use Conda for package management
- Setup an RVSS conda environment
- Download Visual Studio Code for code development
Before you attend the workshop, you will need a copy of this repository on your personal laptop. Clone the repository into your chosen folder on your personal laptop with this terminal command:
git clone --recurse-submodules https://github.com/rvss-australia/RVSS_Need4Speed.git
Navigate to the huggingface dataset and download data.zip. Place this inside your cloned repository and extract the zip file.
You should have a data
folder that contains a train_starter
and val_starter
folder, containing 793 and 436 images respectively.
In this workshop, we will use Conda. Conda is a package manager for Windows, Mac and Linux - it allows you to install packages similar to apt
, homebrew
and vcpkg
. Conda not only supports Python packages, but also packages in C/C++, FORTRAN, and much more.
Nothing that can be done with conda cannot be achieved otherwise. However, with conda it is usually easier and cross-platform.
Conda handles dependencies seamlessly and makes it easy to set up different environments with different versions of libraries.
Conda is also an environment manager (like virtualenv
). Therefore, if an environment is ruined beyond repair, you can just remove it and start over with a clean one.
- We strongly recommend using the community-driven
conda-forge
channel, instead of thedefaults
channel that is maintained by Anaconda. - The easiest way to install conda with conda-forge as default channel is with
Miniforge3
. You can download the installers from here. - If you are on Linux/MacOS, simply type
sh Miniforge3-*.sh
and follow the instructions. On Windows, double clickMiniforge3-Windows-x86_64.exe
and follow the instructions.
- Linux/MacOS: Create a new environment called
rvss
with all required packages:mamba create -n rvss numpy scipy pytorch scikit-learn ipython scikit-image matplotlib tqdm roboticstoolbox-python git ipykernel mediapy py-opencv seaborn gym jupyter spatialmath-python machinevision-toolbox-python ipywidgets plotly torchvision conda-pack tensorboardx pynput click ipympl -c conda-forge --override-channels
. - Windows: Create a new environment called
rvss
with all required packages:mamba create -n rvss numpy scipy pytorch pytorch-cuda scikit-learn ipython scikit-image matplotlib tqdm roboticstoolbox-python git ipykernel mediapy py-opencv seaborn gym jupyter spatialmath-python machinevision-toolbox-python ipywidgets plotly torchvision conda-pack tensorboardx pynput click ipympl -c pytorch -c conda-forge -c nvidia
. - Linux/Windows with GPU (optional as this should happen automatically): Replace
pytorch
withpytorch-gpu
above to enforce a GPU version ofpytorch
. - Windows: You may need to enable long path lengths if you get an error when creating the environment.
- Activating the environment you just created:
conda activate rvss
- Deactivating:
conda deactivate rvss
- Deleting an environment:
conda remove --name FAILED_ENVIRONMENT --all
We strongly recommend downloading Visual Studio Code to use as your code editor during the workshop. You can download VS Code here. Please install at least the following two extensions: Python and Remote - SSH.