/SCW

To be Critical: Self-Calibrated Weakly Supervised Learning for Salient Object Detection

Primary LanguagePython

SCW

Source code for the Paper: "To be Critical: Self-Calibrated Weakly Supervised Learning for Salient Object Detection. "

Jian Wang, Miao Zhang, Yongri Piao, ZhengXuan Ma and HuChuan Lu. IIAU-OIP Lab.

The paper is under review, we will release the PDF upon accepted.











Prerequisites

environment

  • CUDA 10.1
  • pytorch 1.7.1
  • python 3.7.11
  • the others can be found in requirements.txt

training data

our proposed DUTS-Cls dataset can be found in here. code: gpt7

DUTS dataset can be found in here, noting that as a weakly-supervised work, our method only use it's RGB images.

puts the two above datasets in .data/, named as .DUTScls-44/ and .DUTS-train/image/, respectively.

testing data

link: https://pan.baidu.com/s/1PBzDP1Hnf3RIvpARmxn2yA. code: oipw

Training & Testing

Training

Run main.py

here you can adjust the training schedule (such as: learning rate, batchsize, the epoch number of each stage, etc.) in this file.

Testing

Run test_code.py

configure the --test_root as the path of your targeted testset.

the evaluation code can be found in here.

Saliency maps & Checkpoint

saliency maps

link: https://pan.baidu.com/s/1neboLDAs55DHsmsEO4mv4w. code: rvb0

checkpoint

link: https://pan.baidu.com/s/1oywOIqKPMRQrfogNsMXcrA. code: kuih

Acknowledge

Thanks to pioneering helpful works:

  • SSSS: Single-stage Semantic Segmentation from Image Labels, CVPR2020, by Nikita Araslanov et al.
  • IRNet: Weakly Supervised Learning of Instance Segmentation with Inter-pixel Relations, CVPR2019, by Jiwoon Ahn et al.

Citation

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

@article{Wang_SCW,
  author    = {Jian Wang, Miao Zhang, Yongri Piao, Zhengxuan Ma and Huchuan Lu},
  title     = {To be Critical: Self-Calibrated Weakly Supervised Learning for Salient Object Detection},
  year      = {2021},
  url       = {https://arxiv.org/abs/2109.01770},
  eprinttype = {arXiv},
  eprint    = {2109.01770},
}

If you have any questions, please contact me by e-mail: jiangnanyimi@163.com.