Implementation for UTEP

Introduction

This repo provides codes of Learning Unbiased Transferability for Domain Adaptation by Uncertainty Modeling(ECCV2022).

This implementation is based on Transfer-Learning-Library

Prerequisites

pip3 install -r requirements.txt

Train

Run the corresponding commands in run.sh for different tasks. For example, run the following for Amazon to Webcam in Office-31

python main.py /data/office31 -d Office31 -s A -t W -a resnet50 --epochs 20 --seed 1 --log logs/dann/Office31_A2W

Benefit from the Transfer-Learning-Library, If you run these tasks for the first time, the datasets will be downloaded to the corresponding path automatically

Test or Analyse

With the phase setted to test or analysis, the code will run in different mode. Specifically, run the following for testing.

python main.py /data/office31 -d Office31 -s A -t W -a resnet50 --epochs 20 --seed 1 --log logs/dann/Office31_A2W --phase test

And run this for analysis

python main.py /data/office31 -d Office31 -s A -t W -a resnet50 --epochs 20 --seed 1 --log logs/dann/Office31_A2W --phase analysis

If you are interested in our work and want to cite it, you can cite it as

@inproceedings{hu2022learning,
  title={Learning Unbiased Transferability for Domain Adaptation by Uncertainty Modeling},
  author={Hu, Jian and Zhong, Haowen and Yang, Fei and Gong, Shaogang and Wu, Guile and Yan, Junchi},
  booktitle={European Conference on Computer Vision},
  pages={223--241},
  year={2022},
  organization={Springer}
}