README.md

(TPAMI 2024) Divide-and-Conquer: Confluent Triple-Flow Network for RGB-T Salient Object Detection

Hao Tang¹  Zechao Li¹  Dong Zhang²  Shengfeng He³  Jinhui Tang¹

¹Nanjing University of Science and Technology, Nanjing, China
²Hong Kong University of Science and Technology, Hong Kong, China
³Singapore Management University, Singapore 

---

Codes are Coming Soon!

Framework

An overview of the proposed Confluent Triple-Flow Network (ConTriNet), which adopts an efficient ``Divide-and-Conquer'' strategy, is presented. ConTriNet comprises three main flows: a modality-complementary flow that predicts a modality-complementary saliency map, and two modality-specific flows that predict RGB- and Thermal-specific saliency maps, respectively.

VT-IMAG Dataset

The primary purpose of the constructed VT-IMAG is to drive the advancement of RGB-T SOD methods and facilitate their deployment in real-world scenarios. For a fair comparison, all models are solely trained on clear data and simple scenes (i.e., training set of VT5000) and evaluated for Zero-shot Robustness on various real-world challenging cases in VT-IMAG. Download Dataset (Google Drive)

The prediction results of existing RGB-T SOD methods on VT-IMAG are now available for download, enabling researchers to easily compare their methods with existing SOTA methods and directly incorporate these results into their studies. Download VT-IMAG Saliency_maps (Google Drive)

Benchmark Datasets

• VT5000 (ArXiv) Download Datasets (Google Drive)
• VT1000 (ArXiv) Download Datasets (Google Drive)
• VT821 (ArXiv) Download Datasets (Google Drive)

Saliency Maps

The prediction results of existing RGB-T SOD methods and our ConTriNet on benchmark datasets are now available for download, enabling researchers to easily compare their methods with existing SOTA methods and directly incorporate these results into their studies.

• Download VT5000 Saliency_maps (Google Drive)
• Download VT1000 Saliency_maps (Google Drive)
• Download VT821 Saliency_maps (Google Drive)

Evaluation

We use this Saliency-Evaluation-Toolbox for evaluating all RGB-T SOD results.

Citation

Please cite our paper if you find the work useful, thanks!

@article{tang2024divide,
  author={Tang, Hao and Li, Zechao and Zhang, Dong and He, Shengfeng and Tang, Jinhui},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, 
  title={Divide-and-Conquer: Confluent Triple-Flow Network for RGB-T Salient Object Detection}, 
  year={2024},
  pages={1-17},
  doi={10.1109/TPAMI.2024.3511621}}