Code for the paper

Code for the paper in CVPR2019, 'Multi-source weak supervision for saliency detection' (download the pdf file)

Results

score/datasets	ECSSD	HKU-IS	PASCALS	SOD	OMRON	DUTS-test	SED1	SED2
max$F_\beta$	.878	.856	.790	.799	.718	.767	.902	.849
MAE	.096	.084	.134	.167	.114	.096	.081	.097

Download result maps: OneDrive / GoogleDrive

Usage

Test

0. Environment: python2.7, pytorch'1.0'

1. Download models and put in the current folder

2. Run

python main.py \
--img_dir 'path/to/images(.jpg)' \
--gt_dir 'path/to/ground-truth(.png)'

Train

Please checkout the other branch of this repo

Citation

@inproceedings{zeng2019multi,
  title={Multi-source weak supervision for saliency detection},
  author={Zeng, Yu and Zhuge, Yunzhi and Lu, Huchuan and Zhang, Lihe and Qian, Mingyang and Yu, Yizhou},
  booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
  year={2019}
}