Tensor-Aligned Invariant Subspace Learning

When Unsupervised Domain Adaptation Meets Tensor Representations

Proc. IEEE International Conference on Computer Vision (ICCV), 2017

By Hao Lu¹, Lei Zhang², Zhiguo Cao¹, Wei Wei², Ke Xian¹, Chunhua Shen³, Anton van den Hengel³

¹Huazhong University of Science and Technology, China

²Northwestern Polytechnical University, China

³The University of Adelaide, Australia

Introduction

This repository contains the implimentation of Naive Tensor Subspace Learning (NTSL) and Tensor-Aligned Invariant Subspace Learning (TAISL) proposed in our ICCV17 paper.

Prerequisites

1. Matlab is required. This repository has been tested on 64-bit Mac OS X Matlab2016a. The code should also be compatible with Windows 10.
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.
3. Tensor Toolbox at: http://www.sandia.gov/~tgkolda/TensorToolbox/index-2.6.html.
4. Matlab code for optimization with orthogonality constraints at: http://optman.blogs.rice.edu.

For your convenience, these toolboxs have already been included in this repository. Please remember to cite corresponding papers/softwares if you use these codes.

Usage

1. run demo.m for a demonstration for the domain adaptation task of D->C.

Citation

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

@inproceedings{Hao2017,
	author = {Hao Lu and Lei Zhang and Zhiguo Cao and Wei Wei and Ke Xian and Chunhua Shen and Anton van den Hengel},
	title = {When Unsupervised Domain Adaptation Meets Tensor Representations},
	booktitle = {Proc. IEEE International Conference on Computer Vision (ICCV)},
	year = {2017}
}