Contour-guided Saliency Detection with Long-range Interactions

1. Introduction

• We explore the role of line drawings in visual searching;
• Propose a robust light-weight saliency detection network.

2. How to use

2.1 Train model

• install:

 numpy
 pytorch 0.4+
 apex (https://github.com/NVIDIA/apex)
 opencv3

• set your paths in train_edgenet_lrsp.py; (including datapath,savepath,ck_imgpath, etc.)
• run train_edgenet_lrsp.py

2.2 Test

• set your paths in test_edgenet_lrsp.py, model is in (.\ckpts\edgenet\salient_rgb_model_sharewgt_50.pth)
• run test_edgenet_lrsp.py

3. Cite this work

  title={Contour-guided saliency detection with long-range interactions},
  author={Peng, Peng and Yang, Kai-Fu and Liang, Si-Qin and Li, Yong-Jie},
  journal={Neurocomputing},
  volume={488},
  pages={345--358},
  year={2022},
  publisher={Elsevier}
}