TwinsTNet -- TIP 2025

The code is coming soon！

• Please cite our paper if you find it useful for your research.

@ARTICLE{10982382,
  author={Lyu, Pengfei and Yu, Xiaosheng and Chi, Jianning and Wu, Hao and Wu, Chengdong and Rajapakse, Jagath C.},
  journal={IEEE Transactions on Image Processing}, 
  title={TwinsTNet: Broad-View Twins Transformer Network for Bi-Modal Salient Object Detection}, 
  year={2025},
  volume={34},
  number={},
  pages={2796-2810},
  publisher={IEEE}
}

Requirement

• Python (3.7.10+), Pytorch (1.7.0+), Cuda, Torchvision, Tensorboard, TensorboardX, Numpy.

Datasets

• RGB-T Datasets: VT821, VT1000, VT5000,, VI-RGBT1500
• RGB-D Datasets: NLPR, NJUD, DUT-RGBD, STERE, SIP, ReDWeb-S.

Pre-trained models

• RGB-T:
• Baidu cloud disk link, fetch code (u9dh)
• RGB-D:
• Training on NLPR and NJUD, Baidu cloud disk link, fetch code (putl)
• Training on NLPR, NJUD and DUT-RGBD, Baidu cloud disk link, fetch code (rlex)

Results

• RGB-T: The RGB-T results can be downloaded here.

	Em	Sm	wF	Fm	MAE
VT821	0.940	0.911	0.865	0.869	0.023
VT1000	0.954	0.943	0.922	0.913	0.015
VT5000	0.950	0.912	0.874	0.880	0.023
VI-RGBT1500	0.956	0.909	0.880	0.895	0.028

• RGB-D: The RGB-D results will be available soon.

	Em	Sm	wF	Fm	MAE
NLPR	0.976	0.941	0.922	0.922	0.016
NJUD	0.940	0.934	0.925	0.935	0.026
DUT-RGBD	0.973	0.948	0.946	0.956	0.018
SIP	0.949	0.915	0.907	0.926	0.034
STERE	0.945	0.925	0.905	0.913	0.029
ReDWeb	0.780	0.756	0.696	0.728	0.109

Evaluation Metrics Toolbox

• The Evaluation Metrics Toolbox is available here.

Acknowledgements

• Thanks to all the seniors who put in the effort.

Contact Us

If you have any questions, please contact us ([email protected]).