berkerdemirel / adrmx

ADRMX: Additive Disentanglement of Domain Features with Remix Loss

ADRMX: Additive Disentanglement of Domain Features with Remix Loss

This is an official implementation of ADRMX: Additive Disentanglement of Domain Features with Remix Loss. In this repository, we provide the codes for required DomainBed extensions to reproduce our results.

Preparing Environment

1. Clone the DomainBed repository: DomainBed
2. Download the datasets and adjust the paths
3. Clone this repository to update the files algorithms.py, hparams_registry.py and add utils.py under domainbed directory

Getting Started

1. Conda (Recommended):

   conda create -n ADRMX
   conda activate ADRMX

2. Pip

   pip install -r requirements.txt

3. Example Training Command:

   cd PROJECT_PATH 
   python -m domainbed.scripts.train --data_dir=./domainbed/data/ --algorithm ADRMX --dataset PACS

@article{demirel2023adrmx,
  title={ADRMX: Additive Disentanglement of Domain Features with Remix Loss},
  author={Demirel, Berker, Aptoula, Erchan and Ozkan, Huseyin},
  journal={arXiv preprint arXiv:2308.06624},
  year={2023}
}