CFPS

This project provides the code and results for 'Constrained fixation point based segmentation via deep neural network', Neurocomputing 2019. Paper link Homepage

Our code is implemented based on the Caffe of Amulet. You can first install and compile the caffe according to the Amulet.

OSIE-CFPS Dataset

We build a new dataset based on OSIE dataset for 'Constrained Fixation Point based Segmentation' task, you can downlown the dataset here.

OSIE-CFPS contains 3,683 images with corresponding fixation density maps and groundtruths, divided into training set (3,075 images) and testing set (608 images).

Results on OSIE-CFPS and GrabCut

We provide results of the compared 7 methods (GraphCut, RandomWalk, GSC, GBOS, SOS, AVS and SegNet) and our method on 2 datasets: OSIE-CFPS and GrabCut.

The results of GrabCut dataset are one of the 4 runs.

Testing

Download the trained model (FDMNet_iter_40000.caffemodel), and put it under models/FDM/.
Run matlab/FDM_test/test_model.m.
Results of OSIE-CFPS are under models/FDM/OSIE/binary_FDM_test/; results of GrabCut are under models/FDM/GrabCut_database/binary_result/.

Related works on this task

(TIP_2021_OLBPNet) Personal Fixations-Based Object Segmentation with Object Localization and Boundary Preservation.

Citation

    @ARTICLE{Li_2019_NEUCOM,
            author = {Li, Gongyang and Liu, Zhi and Shi, Ran and Wei, Weijie},
            title = {Constrained fixation point based segmentation via deep neural network},
            journal = {Neurocomputing},
            year = {2019},
            volume = {368},
            pages = {180-187},
            month = {Nov.},}

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