Tune it the Right Way: Unsupervised Validation of Domain Adaptationvia Soft Neighborhood Density (ICCV 2021)

[Project Page] [Paper]

Kuniaki Saito, Donghyun Kim, Piotr Teterwak, Stan Sclaroff, Trevor Darrell, and Kate Saenko

Introduction

In this work, we aim to build a criterion that can tune the hyper-parameters of unsupervised domain adaptation model in an unsupervised way. This repository contains codes used for experiments of image classification, semantic segmentation, and toy datasets.

Directories

We split the codes into four directories, base, cdan, adaptseg, and advent since each method employs different structures.

nc_ps: image classification with pseudo-labeling (PS) and neighborhood clustering (NC), and toy experiments.
cdan: image classification with CDAN and MCC (borrowed from CDAN and MCC).
AdaptSegNet: semantic segmentation with adaptsegnet.
Advent(coming soon) will be from ADVENT.

Requirement

Python 3.6.9, Pytorch 1.6.0, Torch Vision 0.7.0, Apex, and sklearn (0.23.2).

In some experiments, we used the nvidia apex library for memory efficient high-speed training.

To track the training details, we also used neptune, but this is optional configuration. Please follow the instructions on each directory for other requirements.

Reference

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

@article{saito2021tune,
  title={Tune it the Right Way: Unsupervised Validation of Domain Adaptation via Soft Neighborhood Density},
  author={Saito, Kuniaki and Kim, Donghyun and Teterwak, Piotr and Sclaroff, Stan and Darrell, Trevor and Saenko, Kate},
  journal={arXiv preprint arXiv:2108.10860},
  year={2021}
}