Butterfly for Wildly Unsupervised Domain Adaptation

This is the official site for the paper "Butterfly: One-step Approach towards Wildly Unsupervised Domain Adaptation" (http://128.84.4.34/pdf/1905.07720). This work is done by

• Dr. Feng Liu (UTS), [email protected]
• Prof. Jie Lu (UTS), [email protected]
• Dr. Bo Han (HKBU/RIKEN), [email protected]
• Dr. Gang Niu (RIKEN), [email protected]
• A/Prof. Guangquan Zhang (UTS), [email protected]
• Prof. Masashi Sugiyama (UTokyo/RIKEN), [email protected].

This paper has been accepted by NeurIPS'19 LTS Workshop and under the second review of IEEE-TPAMI.

Software version

TensorFlow version is 1.14.0. Python version is 3.7.3. CUDA version is 11.2.

These python files require some basic scientific computing python packages, e.g., numpy. I recommend users to install python via Anaconda (python 3.7.3), which can be downloaded from https://www.anaconda.com/distribution/#download-section . If you have installed Anaconda, then you do not need to worry about these basic packages.

After you install anaconda, tensorflow and download MNIST and SYND data below, you can run codes successfully. Good luck!

Data download

You can download MNIST and SYND data used in our paper from https://drive.google.com/file/d/1f4sepb0JXeSSftfAWd6KPa_AVoWA28rj/view?usp=sharing.

You can download Amazon and BCIS data used in our paper from https://drive.google.com/file/d/1hSXn8ctm2vMHu_iIhVS_xB4cq2JV34na/view?usp=sharing.

Ensure that you have the following folder structure after downloading:

'./data/synth_train_32x32.mat'

'./data/synth_test_32x32.mat'

'./data/train_mnist_32x32.npy'

'./data/test_mnist_32x32.npy'

'./data/amazon.mat'

'./data/dense_setup_decaf7/dense_bing_decaf7.mat'

'./data/dense_setup_decaf7/dense_caltech256_decaf7.mat'

'./data/dense_setup_decaf7/dense_imagenet_decaf7.mat'

'./data/dense_setup_decaf7/dense_sun_decaf7.mat'

Flying Butterfly

There are 8 tasks in Butterfly_total.py file:

1. MNIST to SYND (M2S) under Pair-flip noise 20% (P20)
2. MNIST to SYND (M2S) under Pair-flip noise 45% (P45)
3. MNIST to SYND (M2S) under Symmetry-flip noise 20% (S20)
4. MNIST to SYND (M2S) under Symmetry-flip noise 45% (S45)
5. SYND to MNIST (S2M) under Pair-flip noise 20% (P20)
6. SYND to MNIST (S2M) under Pair-flip noise 45% (P45)
7. SYND to MNIST (S2M) under Symmetry-flip noise 45% (S45)
8. SYND to MNIST (S2M) under Symmetry-flip noise 45% (S45)

You can specify one task and run Butterfly using

python Butterfly_total.py --Task_type 1

--> get results for the first task.

There are 24 tasks in ButterflyNet_Amazon_data.py file. You can run

python ButterflyNet_Amazon_data.py

--> get results for Amazon dataset.

There are 3 tasks in ButterflyNet_Obj_data.py file. You can run

python ButterflyNet_Obj_data.py

--> get results for BCIS dataset.

Citation

If you are using this code for your own researching, please consider citing

@inproceedings{liu2019butterfly,
  title={Butterfly: A Panacea for All Difficulties in Wildly Unsupervised Domain Adaptation},
  author={Liu, F. and Lu, J. and Han, B. and Niu, G. and Zhang, G. and Sugiyama, M.},
  booktitle={NeurIPS LTS Workshop},
  year={2019}
}

Acknowledgment

FL, JL and GZ were supported by the Australian Research Council (ARC) under FL190100149. BH was supported by the RGC Early Career Scheme No. 22200720 and NSFC Young Scientists Fund No. 62006202, HKBU Tier-1 Start-up Grant, HKBU CSD Start-up Grant, HKBU CSD Departmental Incentive Grant, and a RIKEN BAIHO Award. GN and MS were supported by JST AIP Acceleration Research Grant Number JPMJCR20U3, Japan. MS was also supported by the Institute for AI and Beyond, UTokyo.