RDA: Robust Domain Adaptation via Fourier Adversarial Attacking

Updates

Paper

RDA: Robust Domain Adaptation via Fourier Adversarial Attacking
Jiaxing Huang, Dayan Guan, Xiao Aoran, Shijian Lu
School of Computer Science Engineering, Nanyang Technological University, Singapore
International Conference on Computer Vision, 2021.

If you find this code/paper useful for your research, please cite our paper:

@article{huang2021rda,
  title={RDA: Robust Domain Adaptation via Fourier Adversarial Attacking},
  author={Huang, Jiaxing and Guan, Dayan and Xiao, Aoran and Lu, Shijian},
  journal={arXiv preprint arXiv:2106.02874},
  year={2021}
}

Abstract

Unsupervised domain adaptation (UDA) involves a supervised loss in a labeled source domain and an unsupervised loss in an unlabeled target domain, which often faces more severe overfitting (than classical supervised learning) as the supervised source loss has clear domain gap and the unsupervised target loss is often noisy due to the lack of annotations. This paper presents RDA, a robust domain adaptation technique that introduces adversarial attacking to mitigate overfitting in UDA. We achieve robust domain adaptation by a novel Fourier adversarial attacking (FAA) method that allows large magnitude of perturbation noises but has minimal modification of image semantics, the former is critical to the effectiveness of its generated adversarial samples due to the existence of domain gaps. Specifically, FAA decomposes images into multiple frequency components (FCs) and generates adversarial samples by just perturbating certain FCs that capture little semantic information. With FAA-generated samples, the training can continue the random walk and drift into an area with a flat loss landscape, leading to more robust domain adaptation. Extensive experiments over multiple domain adaptation tasks show that RDA can work with different computer vision tasks with superior performance.

Installation

  1. Conda enviroment:

conda create -n rda python=3.6
conda activate rda
conda install -c menpo opencv
pip install torch==1.0.0 torchvision==0.2.1
  1. Clone the ADVENT:

git clone https://github.com/valeoai/ADVENT.git
pip install -e ./ADVENT
  1. Clone the CRST:

git clone https://github.com/yzou2/CRST.git
pip install packaging h5py
  1. Clone the repo:

https://github.com/jxhuang0508/RDA.git
pip install -e ./RDA
cp RDA/crst/*py CRST
cp RDA/crst/deeplab/*py CRST/deeplab

Prepare Dataset

  • GTA5: Please follow the instructions here to download images and semantic segmentation annotations. The GTA5 dataset directory should have this basic structure:

RDA/data/GTA5/                               % GTA dataset root
RDA/data/GTA5/images/                        % GTA images
RDA/data/GTA5/labels/                        % Semantic segmentation labels
...
  • Cityscapes: Please follow the instructions in Cityscape to download the images and validation ground-truths. The Cityscapes dataset directory should have this basic structure:

RDA/data/Cityscapes/                         % Cityscapes dataset root
RDA/data/Cityscapes/leftImg8bit              % Cityscapes images
RDA/data/Cityscapes/leftImg8bit/val
RDA/data/Cityscapes/gtFine                   % Semantic segmentation labels
RDA/data/Cityscapes/gtFine/val
...

Pre-trained models

Pre-trained models can be downloaded here and put in RDA/pretrained_models

Evaluation

To evaluate RDA_FAA_T:

cd RDA/CRST
python evaluate_advent.py --test-flipping --data-dir ../RDA/data/Cityscapes --restore-from ../RDA/pretrained_models/model_FAA_T.pth --save ../RDA/experiments/GTA2Cityscapes_RDA

To evaluate RDA_FAA_S_T:

cd RDA/CRST
python evaluate_advent.py --test-flipping --data-dir ../RDA/data/Cityscapes --restore-from ../RDA/pretrained_models/model_FAA_S_T.pth.pth --save ../RDA/experiments/GTA2Cityscapes_RDA

Training

To train RDA_FAA_T:

cd RDA/rda/scripts
python train.py --cfg configs/RDA.yml

To test RDA_FAA_T:

cd RDA/CRST
./test_best.sh

Acknowledgements

This codebase is heavily borrowed from ADVENT and CRST.

Contact

If you have any questions, please contact: [email protected]

GitHub

https://github.com/jxhuang0508/RDA