/MCL

Multi-level Consistency Learning for Semi-supervised Domain Adaptation, IJCAI 2022

Primary LanguagePython

Multi-level Consistency Learning for Semi-supervised Domain Adaptation

Zizheng Yan, Yushuang Wu, Guanbin Li, Yipeng Qin, Xiaoguang Han, Shuguang Cui, "Multi-level Consistency Learning for Semi-supervised Domain Adaptation", IJCAI 2022. [paper]

1. Requirements

pip install -r requirements.txt

The project is developed with PyTorch and POT. It should work with different versions.

2. Data Preparation

For Domainnet, please follow MME to prepare the data. The expected dataset path pattern is like your-domainnet-data-root/domain-name/class-name/images.png.

For Office-Home, please download the resized images and extract, you will get a .pkl and a .npy file, then specify their paths at line 205 in loader/office_home.py. As the dataset scale of office-home is small, we resize the images to 256x256 and save to a single array so that the data loading is faster.

3. Training

Specify the dataset paths and domains in train.sh, and

bash train.sh

4. Acknowledgement

The code is partly based on MME

5. Citation

@article{yan2022multi,
  title={Multi-level consistency learning for semi-supervised domain adaptation},
  author={Yan, Zizheng and Wu, Yushuang and Li, Guanbin and Qin, Yipeng and Han, Xiaoguang and Cui, Shuguang},
  journal={IJCAI},
  year={2022}
}