I2Attack

An implementation for "Indirect Invisible Poisoning Attacks on Domain Adaptation" (KDD'21) [Paper].

Environment Requirements

The code has been tested under Python 3.6.5. The required packages are as follows:

• numpy==1.18.1
• sklearn==0.22.1
• scikit-image==0.16.2
• Pillow==7.0.0
• torch===1.4.0
• torchvision===0.5.0

Data sets

We used the following data sets in our experiments:

• MNIST, USPS, SVHN
• Office-31, Office-Caltech10, Office-Home
• Image-CLEF, VisDA2017

Run the Codes

For I2Attack on unsupervised domain adaptation on digital data (e.g., svhn and mnist), please run

python train_svhn2mnist.py

For I2Attack on unsupervised domain adaptation on real-world image data (e.g., office-31), please run

pyhton main.py

Acknowledgement

This is the latest source code of I2Attack for KDD2021. If you find that it is helpful for your research, please consider to cite our paper:

@inproceedings{wu2021indirect,
  title={Indirect Invisible Poisoning Attacks on Domain Adaptation},
  author={Wu, Jun and He, Jingrui},
  booktitle={Proceedings of the 27th ACM SIGKDD International Conference on Knowledge Discovery \& Data Mining},
  year={2021},
  organization={ACM}
}