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

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 .