Noise-Sensitive Adversarial Learning for Weakly Supervised Salient Object Detection

---

Accepted paper in IEEE Trans on Multimedia, 'Noise-Sensitive Adversarial Learning for Weakly Supervised Salient Object Detection', Yongri Piao, Wei Wu, Miao Zhang, Yongyao Jiang and Huchuan Lu.

Prerequisites

Environment

• Windows 10
• Torch 1.4.0
• CUDA 11.1
• Python 3.6.5

Training data

link: https://pan.baidu.com/s/1n4YGVRhNabM5td4et9o5sw code: wnvl

Training

1st training stage

Case1 : Update soon

Case2 : We upload our 1st pseudo labels in Training data, you can directly use our offered <stage1_training_map> as pseudo labels for convenience.

2nd training stage

setting the training data to the proper root as follows:

NSALWSS -- datasets -- DUTS_pseudo -- DUTS-TR-Image -- 10553 samples
                
                                   -- stage1_training_map -- 10553 pseudo labels
                
                                   -- stage2_training_map -- 10553 pseudo labels (not necessary but we also offered stage2's pseudo labels for convenience)

training

Run train.py

testing

Run test_code.py You need to configure your desired testset in --test_root

The evaluation code can be found in here.

Saliency maps & Checkpoint

We offer our saliency maps and checkpoints.

Saliency maps

link: https://pan.baidu.com/s/1Dyhy107oQTow1UN1Wg9-KA code: 1kfn

Checkpoints

link: https://pan.baidu.com/s/1aMwHkQb-9C2YmM_P-j8f-A code: 32fi

Contact me

If you have any questions, please contact me: [1157008667@qq.com].

Citation

We really hope this repo can contribute the conmunity, and if you find this work useful, please use the following citation:

@ARTICLE{9716868,
  author={Piao, Yongri and Wu, Wei and Zhang, Miao and Jiang, Yongyao and Lu, Huchuan},
  journal={IEEE Transactions on Multimedia}, 
  title={Noise-Sensitive Adversarial Learning for Weakly Supervised Salient Object Detection}, 
  year={2022},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/TMM.2022.3152567}}