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}
}