Global-prior-guided fusion network for salient object detection

Introduction

• Network:
  • global fusion
  • global cross modality fusion

Dependence

• ubuntu
• pytorch 1.7 (other versions may also work)
• numpy, opencv3, tqdm
• apex for efficient training

How to train

1. prepare your training datasets:
   • we followed A2dele and adopted their training dataset.
   • Download Link; Fetch Code: 0fj8
2. set your paths in train_cross.py
3. run train_cross.py

How to test

1. prepare your test datasets. (Most of them are collected in RGBD-SOD survey)
2. download the model from baiduyun,(fetch code: peng) and put in the root dir
3. run test_cross.py.

How to visualize feature maps

1. put your images and depth images into vis_feature.test_data.
2. set the VisualizationDemo.py to get your intersted feature maps.
3. enjoy!

Results

• Baduyun Result Link; Fetch code: peng
• 百度网盘链接：https://pan.baidu.com/s/1edLCHGd9RA6YyJa8vuYAng . 提取码：peng

Evaluation

• Recommend SOD_Evaluation_Metrics

Acknowledgement

• our website
• SOD_Evaluation_Metrics
• A2dele
• RGBD-SOD survey