xrosliang/caffe-sal

Amulet and UCF methods for Saliency Detection

                                  IIAU (2017) Saliency Detection Models

Introduction

This file contains our saliency detection models appeared in ICCV2017. We delve into the convolutional features in pre-trained FCN models and propose the following methods.

1. Amulet

Title: Amulet: Aggregating Multi-level Convolutional Features for Salient Object Detection

Authors: Pingping Zhang, Dong Wang, Huchuan Lu*, Hongyu Wang and Xiang Ruan

Paper Links:

Google Drive Link: https://drive.google.com/file/d/0B5t8yubOMmTKUnZUYTY5NDdWcTg/view?usp=sharing

One Drive Link: https://1drv.ms/u/s!AnzjJimFPB2Ibqy-4-sxZOHnb2s

Baidu Pan Link: https://pan.baidu.com/s/1gfKkIwN

2. UCF

Title: Learning Uncertain Convolutional Features for Accurate Saliency Detection

Authors: Pingping Zhang, Dong Wang, Huchuan Lu*, Hongyu Wang and Baocai Yin

Paper Links:

Google Drive Link: https://drive.google.com/file/d/0B5t8yubOMmTKbUFUQ1gtekVnNEE/view?usp=sharing

One Drive Link: https://1drv.ms/u/s!AnzjJimFPB2IbY_bYGa-d5HemFw

Baidu Pan Link: https://pan.baidu.com/s/1skCi4Dv

3. SRM

Title: A Stagewise Refinement Model for Detecting Salient Objects in Images

Authors: Tiantian Wang, Ali Borji,Lihe Zhang,Pingping Zhang, Huchuan Lu*

Paper Links:

Google Drive Link: https://drive.google.com/file/d/0B_MpGgTntG47b18yUlJxOGtuNDA/view

One Drive Link: https://1drv.ms/u/s!AnzjJimFPB2IbY_bYGa-d5HemFw

Baidu Pan Link: https://pan.baidu.com/s/1skCi4Dv

How to use

Prerequisites:

Download source code from

Google Drive Link: https://drive.google.com/open?id=0B5t8yubOMmTKazhHdV8yR19tek0

One Drive Link: https://1drv.ms/u/s!AnzjJimFPB2Ib_zU3E15C6AxkXQ

Baidu Pan Link: http://pan.baidu.com/s/1bTAYNc

Follow the official websites of the Caffe framework and install the whole toolbox (necessary Matlab wrappers)

1. Training

The training code is in ./models/Amulet and ./models/UCF. For saving memory, we use the BN method implemented by Alex Kendall etal http://mi.eng.cam.ac.uk/projects/segnet/. The training is followed as examples in Caffe. We use the ImageData layer to load training image pairs. The data list can be generated by Matlab in

Google Drive Link: https://drive.google.com/open?id=0B5t8yubOMmTKblFZajBuTTYyanc

Baidu Pan Link: http://pan.baidu.com/s/1pKXnBdD

The only need is changing the path of data files.

2. Testing

The testing code is in ./matlab/Amulet_test and ./matlab/UCF_test.

For saliency testing,

(1) Get prediction : test_saliency_dataset.m

(2) Get PR value : PR.m

(3) Plot PR curves: code_pr.m

(4) Plot Bar figures: code_bar.m

(5) Get MAE for each method : getmae.m

Note that we have provided the PR curves in ./PR_curves

Results

The saliency maps can be download via baidu yun at http://pan.baidu.com/s/1o7O3lQq.

Citation

Please cite the following papers if our models help your research:

@article{zhang2017amulet,

title={Amulet: Aggregating Multi-level Convolutional Features for Salient Object Detection},

author={Zhang, Pingping and Wang, Dong and Lu, Huchuan and Wang, Hongyu and Ruan, Xiang},

journal={arXiv preprint arXiv:1708.02001},

year={2017}

}

@InProceedings{Zhang_2017_ICCV,

author = {Zhang, Pingping and Wang, Dong and Lu, Huchuan and Wang, Hongyu and Ruan, Xiang},

title = {Amulet: Aggregating Multi-Level Convolutional Features for Salient Object Detection},

booktitle = {The IEEE International Conference on Computer Vision (ICCV)},

month = {Oct},

year = {2017}

}

@article{zhang2017learning,

title={Learning Uncertain Convolutional Features for Accurate Saliency Detection},

author={Zhang, Pingping and Wang, Dong and Lu, Huchuan and Wang, Hongyu and Yin, Baocai},

journal={arXiv preprint arXiv:1708.02031},

year={2017}

}

@InProceedings{Zhang_2017_ICCV,

author = {Zhang, Pingping and Wang, Dong and Lu, Huchuan and Wang, Hongyu and Yin, Baocai},

title = {Learning Uncertain Convolutional Features for Accurate Saliency Detection},

booktitle = {The IEEE International Conference on Computer Vision (ICCV)},

month = {Oct},

year = {2017}

}

@InProceedings{Wang_2017_ICCV,

author = {Wang, Tiantian and Borji, Ali and Zhang, Lihe and Zhang, Pingping and Lu, Huchuan},

title = {A Stagewise Refinement Model for Detecting Salient Objects in Images},

booktitle = {The IEEE International Conference on Computer Vision (ICCV)},

month = {Oct},

year = {2017}

}

Question and connection

If any question, please connect

[email protected]; [email protected]