Code and dataset repository for our AAAI 2025 paper: "Alignment-Free RGB-T Salient Object Detection: A Large-scale Dataset and Progressive Correlation Network".
📄 arXiv: https://arxiv.org/pdf/2412.14576
The model and results are available now. [17th, Jul, 2025]
Thank you for your attention.
✨ Update (2026-01): Google Drive links added (recommended for international users).
- 📦 Dataset (UVT20K compressed): 👉 [Google Drive]
- 📌 Results & checkpoints (full mirror): 👉 [Google Drive]
The compressed UVT20K dataset contains annotations of saliency maps, edges, scribbles, and challenge attributes.
Download here:
- 📦 [Baidu Pan] (code:
v2rc) - 🌍 [Google Drive] (recommended)
✨ Google Drive mirror (recommended for international users):
All released results/checkpoints (same content as the Baidu Pan links below):
👉 [Google Drive]
- 📌 Predicted results (ours): [Baidu Pan] (code:
eekm) - 🧩 Model checkpoints: [Baidu Pan] (code:
gvvw) - 📊 Predicted results (compared methods): [Baidu Pan] (code:
6qqn)
- 📦 Download UVT20K for training and testing (see Dataset section above).
- 🧠 Download the pretrained backbone parameters:
- 📦 [Baidu Pan] (code:
3ifw)
- 📦 [Baidu Pan] (code:
- 🧩 Download the pretrained parameters of IHN from: [IHN].
- 📁 Organize dataset and pretrained model directories.
- 🗂️ Create directories for experiments and checkpoints.
- 🧪 Install PyTorch via
conda:torch==1.12.0,torchvision==0.13.0. - 📦 Install other packages:
pip install -r requirements.txt. - 🔧 Set dataset paths in
./options.py.
python -m torch.distributed.launch --nproc_per_node=2 --master_port=2212 train_parallel.py
python test_produce_maps.py
If you think our work is helpful, please cite:
@inproceedings{wang2025alignment,
title={Alignment-Free RGB-T Salient Object Detection: A Large-scale Dataset and Progressive Correlation Network},
author={Wang, Kunpeng and Chen, Keke and Li, Chenglong and Tu, Zhengzheng and Luo, Bin},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
volume={39},
number={7},
pages={7780--7788},
year={2025}
}
This project is based on the following resources:
📮 For questions or feedback, feel free to email: [email protected]