This package has the source code for the paper "Kernelized Subspace Ranking for Saliency Detection" (ECCV16).
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}
}
- Install prerequsites for
Caffe
(see: Caffe installation instructions) - Compile the
./sds_eccv2014-master/extern/caffe
submodule. - Compile the
./gop_1.3
submodule.
Download pretrained SDS model from 1. Then put it into the ./sds_eccv2014-master
folder.
- Run
demo.m
in./code_superpixels
folder to generate superpixels in a Windows environment. - Train: run
train.m
to generate trained model in the./trained_model
folder. - Test: run
test.m
to generate saliency maps in the./saliency_map
folder.
Download our trained models from https://www.researchgate.net/ (search for this paper). Then put it into the ./trained_model
folder.
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/..
[1] Hariharan B, Arbelaez P, Girshick R, et al. Simultaneous detection and segmentation, ECCV2014