QSFNet

This repo is an official implementation of the QSFNet. Quality-aware Selective Fusion Network for VDT Salient Object Detection. IEEE Transactions on Image Processing (2024).

Prerequisites

Usage

1. Clone the repository

2. Training

Download the pretrained model swin_base_patch4_window12_384_22k.pth and resnet34-333f7ec4.pth.

You can train the three stages entirely by using

python Train_all.py

or train the three stages step by step, using

python Mtrain.py
python QAtrain.py
python Ttrain.py

3. Testing

python test_all.py

4. Evaluation

• We provide saliency maps (fetch code: j9ko) of our QSFNet on VDT-2048 dataset.
• We also provide the saliency maps (fetch code: gu22) of other comparison models in our paper on VDT-2048 dataset.
• We provide the saliency maps of challenging sub-datasets generated by our model and other models cited in our paper.
• The edge Ground Truth of the training set of VDT-2048 dataset can be download here (fetch code: u450)

Citation

@article{bao2024quality,
  title={Quality-aware Selective Fusion Network for VDT Salient Object Detection},
  author={Bao, Liuxin and Zhou, Xiaofei and Lu, Xiankai and Sun, Yaoqi and Yin, Haibing and Hu, Zhenghui and Zhang, Jiyong and Yan, Chenggang},
  journal={IEEE Transactions on Image Processing},
  year={2024},
  volume={33},
  pages={3212 - 3226},
  publisher={IEEE}
}

• If you have any questions, feel free to contact me via: lxbao@hdu.edu.cn or zxforchid@outlook.com.