Skip to content

(TIP'18) An Embarrassingly Simple Approach to Visual Domain Adaptation

License

Notifications You must be signed in to change notification settings

poppinace/ldada

Repository files navigation

LDA-inspired Domain Adaptation

An Embarrassingly Simple Approach to Visual Domain Adaptation

IEEE Transactions on Image Processing, 2018

By Hao Lu1, Chunhua Shen2, Zhiguo Cao1, Yang Xiao1, Anton van den Hengel2

1Huazhong University of Science and Technology, China

2The University of Adelaide, Australia

Introduction

This repository includes the implimentation of LDA-inspired Domain Adaptation (LDADA) proposed in our TIP paper. LDADA can achieve high-quality domain adaptation without explicit adaptation. It is conceptually simple, effective, robust, fast, parameter-free, and applicable to both unsupervised and semi-supervised DA settings.

Prerequisites

  1. Matlab is required. This repository has been tested on 64-bit Mac OS X Matlab2016a and on 64-bit Window 10 Matlab2017a.
  2. LibLinear toolbox at: https://www.csie.ntu.edu.tw/~cjlin/liblinear/. Please remember to install it following the instruction on the website, especially for Windows and Ubuntun users.

Usage

  1. choose your options in the paramInit.m function (e.g., setting opt.dataset to specify a dataset or opt.nclasstrain to specify the experimental protocol).
  2. run batchOfficeCaltech10.m/batchOffice.m/batchSatelliteScene5.m/batchMTFS3.m to reproduce the results of a corresponding dataset.

Citation

If you use our codes in your research, please cite:

@article{lu2018ldada,
  title={An Embarrassingly Simple Approach to Visual Domain Adaptation},
  author={Lu, Hao and Shen, Chunhua and Cao, Zhiguo and Xiao, Yang and van den Hengel, Anton},
  journal={IEEE Transactions on Image Processing},
  volume={27},
  number={7},
  pages={3403--3417},
  year={2018},
  publisher={IEEE}
}

About

(TIP'18) An Embarrassingly Simple Approach to Visual Domain Adaptation

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published