AdverX-Ray

Official repository for AdverX-Ray, powered by the GenerativeZoo.

This work is part of the Xecs TASTI project, nr. 2022005.

Prerequisites

You will need:

• python (see pyproject.toml for full version)
• Git
• Make
• a .secrets file with the required secrets and credentials
• load environment variables from .env
• CUDA >= 12.1
• Weights & Biases account

Installation

Clone this repository (requires git ssh keys)

git clone --recursive git@github.com:caetas/AdverX.git
cd adverx

Install conda environment

conda env create -f environment.yml
conda activate python3.10

On Linux

And then setup all virtualenv using make file recipe

(python3.10) $ make setup-all

You might be required to run the following command once to setup the automatic activation of the conda environment and the virtualenv:

direnv allow

Feel free to edit the .envrc file if you prefer to activate the environments manually.

On Windows

You can setup the virtualenv by running the following commands:

python -m venv .venv-dev
.venv-dev/Scripts/Activate.ps1
python -m pip install --upgrade pip setuptools
python -m pip install -r requirements/requirements.txt

To run the code please remember to always activate both environments:

conda activate python3.10
.venv-dev/Scripts/Activate.ps1

Tracking

The code examples are setup to use Weights & Biases as a tool to track your training runs. Please refer to the full documentation if required or follow the following steps:

1. Create an account in Weights & Biases

2. If you have installed the requirements you can skip this step. If not, activate the conda environment and the virtualenv and run:

   pip install wandb

3. Run the following command and insert you API key when prompted:

   wandb login

Dataset

AdverX utilizes the first iteration of the BIMCV-COVID-19+ dataset, that can be downloaded here. However, the dataset needs to be re-organized to a format that is compatible with the analysis performed in this work. Please follow these steps.

1. Download the first iteration of the dataset (~70 GB) available here.

2. Unzip the folder and move it to data/raw.

3. Untar all the tar.gz files in the folder.

4. Run the script available in src/adverx:

   python preprocess_dataset.py

5. You can delete the original dataset folder in data/raw.

Pretrained Checkpoints

The checkpoints can be downloaded here and should be unzipped and moved to models.

Train the Models

Please run the following commands to train the models in a specific machine.

AdverX-Ray

python train.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --n_epochs 200 --lr 3e-4 --patches_image 32 --in_machine siemens
python train.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --n_epochs 200 --lr 3e-4 --patches_image 36 --in_machine konica
python train.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --n_epochs 200 --lr 3e-4 --patches_image 40 --in_machine philips
python train.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --n_epochs 200 --lr 3e-4 --patches_image 42 --in_machine gmm
python train.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --n_epochs 200 --lr 3e-4 --patches_image 42 --in_machine ge

GLOW

python train_GLOW.py --L 3 --K 32 --hidden_channels 512 --n_bits 16 --lr 1.5e-4 --n_epochs 100 --learn_top --patches_image 32 --in_machine siemens
python train_GLOW.py --L 3 --K 32 --hidden_channels 512 --n_bits 16 --lr 1.5e-4 --n_epochs 100 --learn_top --patches_image 36 --in_machine konica
python train_GLOW.py --L 3 --K 32 --hidden_channels 512 --n_bits 16 --lr 1.5e-4 --n_epochs 100 --learn_top --patches_image 40 --in_machine philips
python train_GLOW.py --L 3 --K 32 --hidden_channels 512 --n_bits 16 --lr 1.5e-4 --n_epochs 100 --learn_top --patches_image 42 --in_machine gmm
python train_GLOW.py --L 3 --K 32 --hidden_channels 512 --n_bits 16 --lr 1.5e-4 --n_epochs 100 --learn_top --patches_image 42 --in_machine ge

VAE

python train_VAE.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --n_epochs 200 --lr 3e-4 --patches_image 32 --in_machine siemens
python train_VAE.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --n_epochs 200 --lr 3e-4 --patches_image 36 --in_machine konica
python train_VAE.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --n_epochs 200 --lr 3e-4 --patches_image 40 --in_machine philips
python train_VAE.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --n_epochs 200 --lr 3e-4 --patches_image 42 --in_machine gmm
python train_VAE.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --n_epochs 200 --lr 3e-4 --patches_image 42 --in_machine ge

Evaluation

We present an example for the Philips machine only, but the process is similar for other machines.

AdverX-Ray

python eval.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --in_machine philips --discriminator_checkpoint ./../../models/AdverX/Discriminator_philips.pt

GLOW

python eval_GLOW.py --L 3 --K 32 --hidden_channels 512 --n_bits 16 --in_machine philips --checkpoint ./../../models/Glow/Glow_32_3_512.pt

VAE

python eval_VAE.py --hidden_dims 64 128 256 512 512 --latent_dim 1024 --in_machine philips --checkpoint ./../../models/VAE/VAE_philips.pt

Documentation

Full documentation is available here: docs/.

Security

Thank you for improving the security of the project, please see the Security Policy for more information.

License

This project is licensed under the terms of the CC-BY-4.0 license. See LICENSE for more details.

Citation

This field will be added soon.