FedCAR: Cross-client Adaptive Re-weighting for Generative Models in Federated Learning

Updating..

Getting Started

Prerequisites

What things you need to install the software and how to install them

conda create -n fedcar python==3.8
conda activate fedcar
conda install pytorch==1.8.0 torchvision==0.9.0 torchaudio==0.8.0 cudatoolkit=11.1 -c pytorch -c conda-forge
or Get inspired by your GPU experience : https://pytorch.org/get-started/previous-versions

Installing

A step by step series of examples that tell you how to get a development env running

Say what the step will be

pip install -r requirements.txt

Train

Data format for stylegan2

python3 dataset_tool.py --source={your original dataset path} --dest={output dir path} --width={image size} --height={image_size}

For conditional training we need label json file

python3 make_labels.py --input_folder={stylegan2 format dataset path} --output={same input_folder path} --task='{normal or abnormal}'

running Training script

CUDA_VISIBLE_DEVICES={gpu_id} train.py --outdir './' --data '{dataset_path}' --batch={batch_size} --cond=True

Built With

• Flower - Federated Learning
• StyleGAN2-ADA - GAN