├── GraphNet .pdf
├── voc_result.png
├── illustration.png
├── context_result.png
└── README.md


/GraphNet .pdf:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/MengyangPu/GraphNet/HEAD/GraphNet .pdf


--------------------------------------------------------------------------------
/voc_result.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/MengyangPu/GraphNet/HEAD/voc_result.png


--------------------------------------------------------------------------------
/illustration.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/MengyangPu/GraphNet/HEAD/illustration.png


--------------------------------------------------------------------------------
/context_result.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/MengyangPu/GraphNet/HEAD/context_result.png


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # GraphNet
 2 | GraphNet: Learning Image Pseudo Annotations for Weakly-Supervised Semantic Segmentation, in ACM Multimedia 2018 (oral) [[Paper]](https://github.com/MengyangPu/GraphNet/blob/master/GraphNet%20.pdf)
 3 | 
 4 | ## Overview
 5 | ![Illustration](illustration.png)
 6 | 
 7 | ### Implementation Details
 8 | The network architecture of [DeepLabv2-VGG16](https://github.com/TheLegendAli/DeepLab-Context2) serves as our network architecture of scribble and bounding box annotations experiments.
 9 | 
10 | ### Results
11 | > #### Results on the PASCAL VOC 2012 dataset
12 |   method                      | annotation    | w/o CRF       | w/ CRF 
13 |   --------------------------- | ------------- | ------------- | -------------
14 |   Ours: ScrGraphNet, Initial  | scribble      |  63.3         |  68.2
15 |   Ours: ScrGraphNet, 1-Round  | scribble      |  64.5         |  68.9
16 |   Ours: BboxGraphNet, Initial | bounding box  |  57.1         |  63.4
17 |   Ours: BboxGraphNet, 1-Round | bounding box  |  61.3         |  65.6
18 |  
19 | ![VOC Segmentation](voc_result.png)
20 | > #### Results on the PASCAL-CONTEXT dataset
21 |   method                      | w/o CRF       | w/ CRF 
22 |   --------------------------- | ------------- | -------------
23 |   Ours: ScrGraphNet, Initial  | 33.1          | 39.7
24 |   Ours: ScrGraphNet, 1-Round  | 33.9          | 40.2
25 |   
26 | ![CONTEXT Segmentation](context_result.png)
27 | 
28 | ## Citation 
29 | If you find the paper useful for your research, please cite:
30 | ```bibtex
31 | @INPROCEEDINGS{Pu2018GraphNet,
32 |     author = {Pu, Mengyang and Huang, Yaping and Guan, Qingji and Zou, Qi},
33 |     title = {GraphNet: Learning Image Pseudo Annotations for Weakly-Supervised Semantic Segmentation},
34 |     booktitle = {ACM Multimedia},
35 |     year = {2018}
36 | }
37 | ```
38 | 


--------------------------------------------------------------------------------