ConceptWhitening

Use concept whitening combined with the ResNet50 model to distinguish bird species using important features.

Team members: Alex Katopodis, Hung Le, Islina Shan Mentor: Prof. Cynthia Rudin, Duke University

Related research: Concept Whitening

Follow the instructions below to set up the environment, prepare the dataset, and run the training script.

Setup

The paths to this repository and the CUB dataset may contain sensitive information, hence they will be supplied to the scripts via a secrets.txt file. In the slurm_scripts directory, create a secrets.txt file in the exact same format as the provided example sample_secrets.txt.

Create the virtual environment and install dependencies

It's recommended to use a virtual environment to install and manage the many Python dependencies in this project.

1. In this directory, run

   python3 -m venv .venv
   source .venv/bin/activate
   pip3 install -r requirements.txt

2. If you have not already done so, set the variable VENV_PATH in secrets.txt to be the absolute path to your virtual environment.

Prepare the CUB dataset for training

1. Download and extract the CUB dataset here. Make sure to move or copy the file attributes.txt inside the attributes directory inside the dataset if it is not already there.
2. Set the variable CUB_PATH in secrets.txt to be the absolute path to your downloaded dataset.
3. In this directory, run

   ./slurm_scripts/make_data.sh

   or, if you are using slurm,

   sbatch slurm_scripts/make_data.sh

Note

Remember to make this script executable by running

sudo chmod +x slurm_scripts/make_data.sh

Training the Model

To train the model, run the following command

./slurm_scripts/train.sh

or, if you are using slurm,

sbatch slurm_scripts/train.sh

Note

Remember to make this script executable by running

sudo chmod +x slurm_scripts/make_data.sh

Note

Ensure you have enough disk space and memory to handle the dataset and model training. The training script will save model checkpoints and logs in the checkpoints/ and logs/ directories, respectively.

You can adjust all the training parameters in config.yaml as needed.