Skip to content

This is a PyTorch implementation of "Cross-modality Discrepant Interaction Network for RGB-D Salient Object Detection" accepted by ACM MM 2021 (poster).

Notifications You must be signed in to change notification settings

zhangchen98/CDINet-ACM-MM21

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

35 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CDINet

This is a PyTorch implementation of "Cross-modality Discrepant Interaction Network for RGB-D Salient Object Detection" accepted by ACM MM 2021 (poster).

Paper: https://dl.acm.org/doi/pdf/10.1145/3474085.3475364.

Arxiv version: https://arxiv.org/pdf/2108.01971.pdf

Network

image

Requirement

Pleasure configure the environment according to the given version:

  • python 3.7.10
  • pytorch 1.8.0
  • cudatoolkit 10.2.89
  • torchvision 0.9.0
  • tensorboardx 2.3
  • opencv-python 4.5.1.48
  • numpy 1.20.2

We also provide ".yaml" files for conda environment configuration, you can download it from [Link], code: 642h, then use conda env create -f CDINet.yaml to create a required environment.

Data Preprocessing

For all depth maps in training and testing datasets, we make a uniform adjustment so that the foreground have higher value than the background, it is very important. Please follow the tips to download the processed datasets and pre-trained model:

  1. Download training data from [Link], code: 0812.
  2. Download testing data from [Link], code: 0812.
  3. Download the parameters of whole model from [Link], code: 0812.
├── backbone 
├── CDINet.pth
├── CDINet_test.py
├── CDINet_train.py
├── dataset
│   ├── CDINet_test_data
│   └── CDINet_train_data
├── model
├── modules
└── setting

Training and Testing

Training command: python CDINet_train.py --gpu_id xx --batchsize xx

You can find the saved models and logs in "./CDINet_cpts".

Testing command: python CDINet_test.py --gpu_id xx

You can find the saliency maps in "./saliency_maps".

Results

  1. Qualitative results: we provide the saliency maps, you can download them from [Link], code: 0812.
  2. Quantitative results:
NLPR NJUD DUT STEREO LFSD
0.9162 0.9215 0.9372 0.9033 0.8746
0.9273 0.9188 0.9274 0.9055 0.8703
0.0240 0.9354 0.0302 0.0410 0.0631

Bibtex

@inproceedings{Zhang2021CDINet, 
    author = {Zhang, Chen and Cong, Runmin and Lin, Qinwei and Ma, Lin and Li Feng and Zhao, Yao and Kwong, Sam},   
    title = {Cross-modality Discrepant Interaction Network for {RGB-D} Salient Object Detection},     
    booktitle = {Proceedings of the 29th ACM International Conference on Multimedia},     
    year = {2021},
    organization={ACM}
} 

Contact

If you have any questions, please contact Chen Zhang at chen.zhang@bjtu.edu.cn .

About

This is a PyTorch implementation of "Cross-modality Discrepant Interaction Network for RGB-D Salient Object Detection" accepted by ACM MM 2021 (poster).

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages