PFOS

This project provides the code and results for 'Personal Fixations-Based Object Segmentation with Object Localization and Boundary Preservation', IEEE TIP 2021. IEEE Link or arxiv Link Homepage

Our code is implemented based on the Caffe of Amulet. You can first install and compile the caffe according to our previous work CFPS or original Amulet.

PFOS Dataset

We build a new dataset based on OSIE dataset for 'Personal Fixations-based Object Segmentation' (PFOS) task, you can download the PFOS dataset here (code: npqn).

PFOS dataset contains 700 images and 10,500 free-view personal fixation maps, each image has 15 personal fixation maps from 15 subjects and the transformed binary groundtruths. It is divided into the training set (600 images with 9,000 free-view personal fixation maps) and the testing set (100 images with 1,500 free-view personal fixation maps).

PFOS task and OLBPNet

PFOS task:

OLBPNet Overview:

Testing

1. Install and compile the caffe according to our previous work CFPS or original Amulet.
2. Download the trained model (code: evk9) (FDMAttBlock_iter_30000.caffemodel), and put it under models/OLBPNet/.
3. Put the PFOS testing set under models/OLBPNet/PFOS/test/.
4. The measure code is under matlab/OLBPNet_test/, run matlab/OLBPNet_test/test_model.m.
5. Results of PFOS testing set are under models/OLBPNet/PFOS/binary_test/.
6. The measure code is under matlab/OLBPNet_test/Measure_all_FDM.m.

Results on PFOS Testing Set

We provide results (code: jios) of the compared 17 methods and our method on PFOS testing set.

Performance

Related works on this task

(NEUCOM_2019_CFPS) Constrained Fixation Point based Segmentation via Deep Neural Network.

Citation

    @ARTICLE{Li_2021_OLBP,
            author = {Gongyang Li and Zhi Liu and Ran Shi and Zheng Hu and Weijie Wei and Yong Wu and Mengke Huang and Haibin Ling},
            title = {Personal Fixations-Based Object Segmentation With Object Localization and Boundary Preservation},
            journal = {IEEE Transactions on Image Processing},
            year = {2021},
            volume = {30},
            pages = {1461-1475},}

If you encounter any problems with the code, want to report bugs, etc.

Please contact me at lllmiemie@163.com or ligongyang@shu.edu.cn.