Universal Multi-Source Domain Adaptation

Code release for Universal Multi-Source Domain Adaptation for Image Classification (Pattern Recognition, 2022, IF: 8.518)

• All rights reserved by Yueming Yin, Email: [email protected] (or [email protected]).

Requirements

• python 3.6+
• PyTorch 1.0

pip install -r requirements.txt

Usage

• Download datasets from https://github.com/jindongwang/transferlearning

• Generate the list of your datasets (read the comments in the code and modify the relevant parameters to use, lists we used in the Dataset_lists folder are as a reference):

  python turn_to_list.py

• Download pre-trained model from https://download.pytorch.org/models/resnet50-19c8e357.pth

• Write your config file in "config.yaml"

• Train (configurations in train-config-office31.yaml are only for Office-31 dataset):

  python main.py --config train-config-office31.yaml

• Test

  python main.py --config test-config-office31.yaml

• Monitor (tensorboard required)

  tensorboard --logdir .

Best Cases

We provide the representative best cases and config files for Office-31 datasets at Google Drive.

Citation

Please cite:

@article{yin2022universal,
  title={Universal multi-Source domain adaptation for image classification},
  author={Yin, Yueming and Yang, Zhen and Hu, Haifeng and Wu, Xiaofu},
  journal={Pattern Recognition},
  volume={121},
  pages={108238},
  year={2022},
  publisher={Elsevier}
}

or

Yin, Yueming, Zhen Yang, Haifeng Hu, and Xiaofu Wu. "Universal multi-Source domain adaptation for image classification." Pattern Recognition 121 (2022): 108238.

Contact

• [email protected] (Yueming Yin)