A Pytorch Implementation of Open Set Domain Adaptation by Back-propagation (ECCV 2018)

Introduction

Official Implementation of Open Set Domain Adaptation by Back-propagation. We publicize code for VisDA experiment.

Data Preparation

Follow https://github.com/VisionLearningGroup/taskcv-2017-public/tree/master/classification to download visda dataset. Input path to utils/list_visda.py and run list_visda.py to generate path list file.

Train

 CUDA_VISIBLE_DEVICES=$GPU_ID python trainer_osda.py \
                    --net alex

Note

We reported the performance using pytorch 0.3. Results below are obtained using pytorch 0.4 for Alexnet and VGG. We are investigating the issues caused by the chanage in versions.

Network	bicycle	bus	car	motorcycle	train	truck	unknown
AlexNet	54.3	76.0	38.6	77.8	71.7	1.0	70.6
VGG	58.4	69.6	50.0	81.3	81.2	28.3	91.7

Citation

Please cite the following reference if you utilize this repository for your project.

@inproceedings{saito2018open,
  title={Open set domain adaptation by backpropagation},
  author={Saito, Kuniaki and Yamamoto, Shohei and Ushiku, Yoshitaka and Harada, Tatsuya},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
  pages={153--168},
  year={2018}
}