This is the code for the paper

Geometric and Textural Augmentation for Domain Gap Reduction

Project Page | Paper | Poster | Video

Preresquisites

Testbed Install:

We use Dassl as the testbed and the code is based on it.

# Create the conda environment (make sure conda is installed)
conda create -n dassl python=3.7
conda activate dassl

# Install dependencies
cd Dassl/
pip install -r requirements.txt

# Install torch (version >= 1.7.1) and torchvision based on your cuda version 
conda install pytorch torchvision cudatoolkit=your_cuda_version -c pytorch

# Install this library (no need to re-build if the source code is modified)
python setup.py develop

Datasets Install:

We use three commonly used multi-domain datasets (please download datasets into this folder):

• PACS (Li et al., 2017) | Download Link: google drive.
• Office-Home-DG (Venkateswara et al., 2017) | Download Link: google drive.
• Digits-DG | Download Link: google drive.

Download the style predictor model into this folder.

Training and Testing

cd gta-dgr/scripts/

# Training on PACS
bash pacs.sh

# Training on Office-Home
bash officehome.sh

# Training on Digits-DG
bash digits.sh

If you find this code useful for your research, please cite

@InProceedings{Liu22GTDG, 
  author={Xiao-Chang Liu and Yong-Liang Yang and Peter Hall},
  title={Geometric and Textural Augmentation for Domain Gap Reduction},
  booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2022}
}