Awesome
Detect Globally, Refine Locally: A Novel Approach to Saliency Detection (DGRL)
This package has the source code for the paper "Detect Globally, Refine Locally: A Novel Approach to Saliency Detection" (CVPR18).
Paper link
- The paper can be found in Baidu drive or Google drive.
How to use
Train
- For Global Localizaiton Network (GLN), using the code in
./stage1/train/for training. The example images are given in./stage1/train/data/. Download the initialized model from Baidu drive or Google drive. - For Boundary Refinement Network (BRN), using the code in
./stage2/train/for training. After finishing the training process of the GLN, then run the code in./stage1/test/test.mto generate saliency maps of GLN. The examples images are given in./stage2/train/data/. Each image in the training set, including the saliency map generated by the GLN, the original RGB image, the ground truth, should be resize to 480 * 480 by the 'nearest' method. Using the code in./stage2/train/init/generate_train.mto generate the initialization model.
Test
- Download our trained model from Baidu drive or Google drive.
- Run
./stage1/test/test.mto generate saliency maps of Global Localizaiton Network (GLN). - Run
./stage2/test/test.mto generate saliency maps of Boundary Refinement Network (BRN). The saliency maps of GLN will serve as the input of BRN.
Download
The saliency maps on 10 datasets including ECSSD, PASCAL-S, SOD, SED1, SED2, MSRA, DUT-OMRON, THUR15K, HKU-IS and DUTS can be found in the following links.
GLN: Baidu drive or Google drive.
BRN: Baidu drive or Google drive.
Cite this work
If you find this work useful in your research, please consider citing:
@inproceedings{wang2018detect,
title={Detect Globally, Refine Locally: A Novel Approach to Saliency Detection},
author={Wang, Tiantian and Zhang, Lihe and Wang, Shuo and Lu, Huchuan and Yang, Gang and Ruan, Xiang and Borji, Ali},
booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
pages={3127--3135},
year={2018}
}