Skip to content

VisionLearningGroup/CDS

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Donghyun Kim, Kuniaki Saito, Tae-Hyun Oh, Bryan A. Plummer, Stan Sclaroff, and Kate Saenko

Overview

Introduction

We present Cross-Domain Self-supervised Pretraining (CDS), a two-stage pre-training approach that improves the generalization ability of standard single-domain pre-training. This is an PyTorch implementation of CDS. This implementation is based on Instance Discrimination, CDAN, and MME. This repository currently supports codes for experiments for few-shot DA on Office-Home. We will keep updating this repository to support other experiments.

Directories

CDS_pretraining: Our implementation of CDS.
data: Datasets (e.g. Office-Home) used in this paper
CDAN: Implementation borrowed from CDAN

Requirement

Python 3.6.9, Pytorch 1.6.0, Torch Vision 0.7.0.

Usage

CDS Pre-training

CDS pretraining on Real and Clipart domains in Office-Home.

cd CDS_pretraining
python CDS_pretraining.py --dataset office_home --source Real --target Clipart

Few-shot DA (1-shot and 3-shots)

cd CDAN
python CDAN_fewshot_DA.py --method CDAN+E --dataset office_home --source Real --target Clipart

Reference

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

@inproceedings{kim2021cds,
  title={CDS: Cross-Domain Self-Supervised Pre-Training},
  author={Kim, Donghyun and Saito, Kuniaki and Oh, Tae-Hyun and Plummer, Bryan A and Sclaroff, Stan and Saenko, Kate},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={9123--9132},
  year={2021}
}

About

CDS: Cross-Domain Self-supervised Pre-training

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages