Home

Awesome

Cross-modal Hierarchical Interaction Network for RGB-D Salient Object Detection

image
Figure.1 The overall architecture of the proposed HINet model.
The paper can be downloaded from here[code:NEPU], which is published in Patern Recognition 🎆.

1.Requirements

Python v3.6, Pytorch 0.4.0+, Cuda 10.0, TensorboardX 2.0, opencv-python

2.Data Preparation

Download the test data from here[code:NEPU], test_in_train data from here[code:NEPU]. Then put them under the following directory:

-Dataset\   
   -train\  
   -test\ 
       -NLPR\
       -STERE\
       -SSD\
       -LFSD\
       -NJUD\
   -test_in_train\
   

3.Training/Testing & Evaluating

Please download the released code and then:

run python Train.py  

Please complete the training or download the pre-trained weights from here[code:NEPU] or Google, and then:

run python Test.py  

Then the test maps will be saved to './Salmaps/'

You can evaluate the result maps using the tool from here[code:NEPU], thanks for Dengpin Fan.

4.Results

image
Figure.2 Qualitative comparison of our proposed method with some SOTA methods.

image
Table.1 Quantitative comparison with some SOTA models on five public RGB-D benchmark datasets.

5.Citation

Thank you for your interest in our work, please cite:

@article{BI2022109194, title = {Cross-modal Hierarchical Interaction Network for RGB-D Salient Object Detection}, journal = {Pattern Recognition}, pages = {109194}, year = {2022}, issn = {0031-3203}, doi = {https://doi.org/10.1016/j.patcog.2022.109194}, url = {https://www.sciencedirect.com/science/article/pii/S0031320322006732}, }

6.Contact

If you have any questions, feel free to contact us via tianzhu.xiang19@gmail.com (T.-Z. Xiang) or wuranwan2020@sina.com (Ranwan Wu). For more related work, you can also visit tianzhu.xiang