DSNet

Dynamic Selective Network for RGB-D Salient Object Detection

This repo is an official implementation of the DSNet, which has been accepted by the journal IEEE Transactions on Image Processing, 2021.

Prerequisites

• python=3.x
• pytorch=1.0.0+
• torchvision
• numpy
• opencv-python

Usage

1. Clone the repository

git clone https://github.com/Brook-Wen/DSNet.git
cd DSNet/

2. Training

python main.py --gpu '0' --lr 1e-5 --batch_size 4

• Training set: NJU2K (1,485), NLPR (700).
• Make sure that the GPU memory is enough.

3. Testing

python main.py --gpu '0' --batch_size 1 --mode='test' --model='[YOUR PATH]' --test_fold='[SAVE PATH]' --sal_mode='[DATASET]'

• We provide the pre-trained model (fetch code: ve9d).
• We evaluate our DSNet on eight commonly used datasets: NJU2K, NLPR, STERE, DES, LFSD, SSD, SIP and ReDWeb-S. These datasets can be downloaded from the links provided in http://dpfan.net/d3netbenchmark/.

4. Evaluation

• We provide saliency maps (fetch code: nd4m) of our DSNet on 8 datasets.
• You can use this toolbox for evaluation.

Citation

If you think this work is helpful, please cite

@inproceedings{wen2021dsnet,
  title={Dynamic Selective Network for RGB-D Salient Object Detection},
  author={Wen, Hongfa and Yan, Chenggang and Zhou, Xiaofei and Cong, Runmin and Sun, Yaoqi and Zheng, Bolun and Zhang, Jiyong and Bao, Yongjun and Ding, Guiguang},
  booktitle={IEEE Transactions on Image Processing},
  year={2021}
}

• If you have any questions, feel free to contact me via: hf_wen(at)outlook.com.