A lightweight, easy-to-extend, easy-to-learn, high-performance, and for-fair-comparison toolkit based on PyTorch for domain adaptation (DA) of deep neural networks.

As initial version, we support the following algoirthms. We are working on more algorithms. Of course, you are welcome to add your algorithms here.

1. DDC (Deep Domain Confusion) [1] / DAN (Deep Adaptation Network) [2]
2. DeepCoral [3]
3. DANN (Domain-adversarial neural network / RevGrad) [4]
4. DSAN (Deep Subdomain Adaptation Network) [5]
5. DAAN (Dynamic Adversarial Adaptation Network) [6]
6. BNM (Batch Nuclear-norm Maximization) [7]

The detailed explanation of those methods are in the README.md files of corresponding methods.

You can either git clone this whole repo by:

git clone https://github.com/jindongwang/transferlearning.git
cd code/DeepDA
pip install -r requirements.txt

Or if you just want to use this folder (i.e., no other things in this big transferlearning repo), you can go to this site and paste the url of this DeepDA folder (https://github.com/jindongwang/transferlearning/edit/master/code/DeepDA) and then download only this folder!

We recommend to use Python 3.7.10 which is our development environment.

1. Modify the configuration file in the corresponding directories
2. Run the main.py with specified config, for example, python main.py --config DAN/DAN.yaml

• We provide shell scripts to help you reproduce our experimental results: bash DAN/DAN.sh.

It is easy to design your own method following the 3 steps:

1. Check whether your method requires new loss functions, if so, add your loss in the loss_funcs
2. Check and write your own model's file inherited from our models.TransferNet
3. Write your own config file (.yaml)

We present results of our implementations on 2 popular benchmarks: Office-31 and Office-Home. We did not perform careful parameter tuning and simply used the default config files. You can easily reproduce our results using provided shell scripts! Note that source-only method is the one that uses no domain adaptation methods as a naive baseline method.

The toolkit is under active development and contributions are welcome! Feel free to submit issues and PRs to ask questions or contribute your code. If you would like to implement new features, please submit a issue to discuss with us first.

[1] Tzeng, Eric, et al. "Deep domain confusion: Maximizing for domain invariance." arXiv preprint arXiv:1412.3474 (2014).

[2] Long, Mingsheng, et al. "Learning transferable features with deep adaptation networks." International conference on machine learning. PMLR, 2015.

[3] Sun, Baochen, and Kate Saenko. "Deep coral: Correlation alignment for deep domain adaptation." European conference on computer vision. Springer, Cham, 2016.

[4] Ganin, Yaroslav, and Victor Lempitsky. "Unsupervised domain adaptation by backpropagation." International conference on machine learning. PMLR, 2015.

[5] Zhu, Yongchun, et al. "Deep subdomain adaptation network for image classification." IEEE transactions on neural networks and learning systems (2020).

[6] Yu, Chaohui, et al. "Transfer learning with dynamic adversarial adaptation network." 2019 IEEE International Conference on Data Mining (ICDM). IEEE, 2019.

[7] Cui, Shuhao, et al. Towards discriminability and diversity: Batch nuclear-norm maximization under label insufficient situations. CVPR 2020.

If you think this toolkit or the results are helpful to you and your research, please cite us!

@Misc{deepda,
howpublished = {\url{https://github.com/jindongwang/transferlearning/tree/master/code/DeepDA}},   
title = {DeepDA: Deep Domain Adaptation Toolkit},  
author = {Wang, Jindong and Hou, Wenxin}
}  

• Wenxin Hou: houwx001@gmail.com
• Jindong Wang: jindongwang@outlook.com