Universal Domain Adaptation through Self-Supervision (NeurlPS 2020)

This repository provides code for the paper, Universal Domain Adaptation through Self-Supervision. Please go to our project page to quickly understand the content of the paper or read our paper.

[Project Page] [Paper (will be updated soon)]

Environment

Python 3.6.9, Pytorch 1.2.0, Torch Vision 0.4, Apex. See requirement.txt. We used the nvidia apex library for memory efficient high-speed training.

Data Preparation

Office Dataset OfficeHome Dataset VisDA

Prepare dataset in data directory as follows.

./data/amazon/images/ ## Office
./data/Real/ ## OfficeHome
./data/visda_train/ ## VisDA synthetic images
./data/visda_val/ ## VisDA real images

Prepare image list.

unzip txt.zip

File list has to be stored in ./txt.

Train

All training script is stored in script directory.

Example: Open Set Domain Adaptation on Office.

sh script/run_office_obda.sh $gpu-id configs/office-train-config_ODA.yaml

Reference

This repository is contributed by Kuniaki Saito. If you consider using this code or its derivatives, please consider citing:

@inproceedings{saito2020dance,
  title={Universal Domain Adaptation through Self-Supervision},
  author={Saito, Kuniaki and Kim, Donghyun and Sclaroff, Stan and Saenko, Kate},
  journal={NeurIPS},
  year={2020}
}