Kernelized Subspace Ranking for Saliency Detection

This package has the source code for the paper "Kernelized Subspace Ranking for Saliency Detection" (ECCV16).

Citing this work

If you find this work useful in your research, please consider citing:

 @inproceedings{wangeccv16,
    Author={Tiantian Wang and Lihe Zhang and Huchuan Lu and Chong Sun and Jinqing Qi},
    Title={Kernelized Subspace Ranking for Saliency Detection},
    Booktitle={European Conference on Computer Vision (ECCV)},
    Year={2016}
 }

Installation

1. Install prerequsites for Caffe (see: Caffe installation instructions)
2. Compile the ./sds_eccv2014-master/extern/caffesubmodule.
3. Compile the ./gop_1.3 submodule.

Prerequisites

Download pretrained SDS model from 1. Then put it into the ./sds_eccv2014-master folder.

Train & Test

1. Run demo.m in ./code_superpixels folder to generate superpixels in a Windows environment.
2. Train: run train.m to generate trained model in the ./trained_model folder.
3. Test: run test.m to generate saliency maps in the ./saliency_map folder.

Our Trained Model

Download our trained models from https://www.researchgate.net/ (search for this paper). Then put it into the ./trained_model folder.

Saliency Map

The zip file of saliency maps on the SED1, SED2, SOD, PASCAL, MSRA, HKU-IS, THUR15K, ECSSD and DUT-OMRON datasets can be downloaded from https://www.researchgate.net/ or https://github.com/..

Contact

tiantianwang.ice@gmail.com

[1] Hariharan B, Arbelaez P, Girshick R, et al. Simultaneous detection and segmentation, ECCV2014