Charlie-Vision

The AI behind Charlie X

Setup

pip

You will required python installed. There are multiple environment files, but for most cases it suffices to use vision-docker/requirements.txt.

First create a virutal environment (we'll call ours vision) and activate it:

python3 -m venv vision
source ./vision/bin/activate

You should see your prompt be prefixed with the name of the environment (vision).

Next install all the required packages:

pip install -r vision-docker/requirements.txt

And that should be all.

conda

conda is another way to manage environments. However many of their dependencies have problems so it might be difficult to use.

The instructions below have worked so far.

To start working on this repository, you will need to have miniconda/ anaconda installed, I recommend installing miniconda.

The list of installers can be foudn on the miniconda docs.

The installation guide was written with reference to their macos install instructions

M1

If you are using a Mac running on M1, you should be able to use the following instructions. First to install miniconda in your $HOME directory, run the following command to download the appropriate setup file:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-MacOSX-arm64.sh -O ~/miniconda.sh
bash ~/miniconda.sh -p $HOME/miniconda

The installer prompts “Do you wish the installer to initialize Miniconda3 by running conda init?” The website and I recommend “yes”.

To verify your install, check that you have conda version 22 or later by running:

conda --version

NUS HPC

Miniconda is stored as a module on NUS HPC. The instructions for accessing it was taken from their Python guide. Simply run the following commands to setup

echo ". /app1/bioinfo/miniconda/4.9/etc/profile.d/conda.sh" >> ~/.bashrc
mkdir ~/conda_envs
echo "export CONDA_ENVS_PATH=~/conda_envs/" >> ~/.bashrc

Now to enable conda run the following:

. ~/.bashrc # This is only needed if you just edited your .bashrc file. A .bashrc file is run automatically everytime you login, so if you restarted terminal again this is unnecessary.
module load miniconda
conda activate base

You should see the (base) prefix added to your prompt.

Development

GitHub

To start working on this repository, first clone it into your workspace

git clone https://github.com/Charlie-X-Ray/Charlie-Vision.git
cd Charlie-Vision

Creating Conda Environment

Conda is an environment and package manager that supposedly helps install packages and check their dependencies.

First start by creating an environment that uses python3.11. This will also give us access to pip:

conda create env -n cxr-pip python=3.11 -y
conda activate cxr-pip

You should see a (cxr-pip) suffix added to your prompt

Next we pip install a bunch of packages

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu121 #this also gives access to PIL
pip install pydicom pandas matplotlib opencv-python

This should be everything you need.

Creating Conda Env From File (WIP)

Note that This is bug prone, so I recommend following the steps above

Run the following to load from a .yml file:

conda env create -f environment-pip.yml

To activate your environment run:

conda activate cxr-pip

You should see a prefix (cxr-pip) added to your prompt.

Test Conda Envrionment

To test that the environment was created correctly run in the Charlie-Vision directory:

python test.py

If some of the imports fail, it means not all the pacakges were installed correctly. Try deleting the created envrionment and repeating the creating conda environment steps.

If this all works, run main.py to draw boxes using the following:

python main.py