STEM: An approach to Multi-source Domain Adaptation with Guarantees

Introduction

This is the official implementation of ``STEM: An approach to Multi-source Domain Adaptation with Guarantees''

Prerequisites

System Requirement:

• Anaconda3
• Cuda toolkit 10.0

Install other environment requirement by Anaconda3 following:

conda env create -f env.yml

Note: the environment requires tensorbayes libs, however, available tensorbayes using Python 2.7. To fix the problem, please download tensorbayes, untar it and override to the env (stem) folder:

tar -xvf tensorbayes.tar
cp -rf ./tensorbayes/* /opt/conda/envs/stem/lib/python3.6/site-packages/tensorbayes/

Dataset Preparation

Please download and unzip the dataset and save all *.mat file under ../datasets. To save time computing, we extracted ResNet101 feature for Office-Caltech10 and provided as following:

• Digits-five
• Office-31
• DomainNet

Training

The config parameter to train model in config folder, please check it before run. To train the model, simply run:

python run_stem_ht_mimic_hs.py --config 
   
    .yaml --trg_name 
    

    
   

For example:

Train Digit-five with target domain is Synthetic Digits:

python run_stem_ht_mimic_hs.py --config DigitFive.yaml --trg_name syn

Train Office-Caltech10 with target domain is Amazon:

python run_stem_ht_mimic_hs.py --config OfficeCaltech10.yaml --trg_name amazon

Cite

Please cite the paper whenever our STEM is used to produce published results or incorporated into other software:

@InProceedings{Nguyen_2021_ICCV,
author    = {Nguyen, Van-Anh and Nguyen, Tuan and Le, Trung and Tran, Quan Hung and Phung, Dinh},
title     = {STEM: An Approach to Multi-Source Domain Adaptation With Guarantees},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month     = {October},
year      = {2021},
pages     = {9352-9363}

}