README.md

CMINet

Cross-stage Multi-scale Interaction Network for RGB-D Salient Object Detection

This is the official implementation of "Cross-stage Multi-scale Interaction Network for RGB-D Salient Object Detection" as well as the follow-ups. The paper has been published by IEEE Signal Processing Letters, 2022. The paper link is here.

---

Content

• Run CMINet code
• Pretrained models
• Saliency maps
• Evaluation tools
• Citation

---

Run CMINet code

• Train
  run python train.py
  # put pretrained models in the pretrained folder
  # set '--train-root' to your training dataset folder

• Test
  run python test.py
  # set '--test-root' to your test dataset folder
  # set '--ckpt' as the correct checkpoint
  

---

Pretrained models

• The pretrained models can be downloaded in Baidu Cloud (fetach code is pcmi). Then put the pretrained models such as 'resnet_50.pth' in the pretrained folder.

---

Saliency maps

• The saliency maps can be approached in Baidu Cloud (fetach code is cmin). Note that all testing results are provided not only including those listed in the paper.

---

Evaluation tools

• The evaluation tools, training and test datasets can be obtained in RGBD-SOD-tools.

---

Citation

@ARTICLE{yi2022cross,
  author={Yi, Kang and Zhu, Jinchao and Guo, Fu and Xu, Jing},
  journal={IEEE Signal Processing Letters}, 
  title={Cross-Stage Multi-Scale Interaction Network for RGB-D Salient Object Detection}, 
  year={2022},
  volume={29},
  number={},
  pages={2402-2406},
  doi={10.1109/LSP.2022.3223599}
}

---