Home

Awesome

SDDNet_ACMMM23

Runmin Cong, Yuchen Guan, Jinpeng Chen, Wei Zhang, Yao Zhao, and Sam Kwong, SDDNet: Style-guided dual-layer disentanglement network for shadow detection, ACM Multimedia (ACM MM), 2023. In Press.

Network

Our overall framework:

image

Requirement

Pleasure configure the environment according to the given version:

We also provide ".yaml" files for conda environment configuration, you can use conda env create -f env.yaml to create a required environment.

ResNext101 has been adopted, please put resnext_101_32x4d.pth in the SDDNet/resnext directory. You can download the model from [Link], code: mvpl.

Preparation

Please follow this structure to inspect the code:

├── ISTD_Dataset
    ├── test
    ├── train
├── SBU-shadow
    ├── SBU-Test_rename
    ├── SBUTrain4KRecoveredSmall
├── UCF
    ├── train_A
    ├── train_B
├── SDDNet
    ├── ckpt
    ├── datasets
    ├── logs
    ├── networks
    ├── resnext
    ├── test
    ├── utils
    ├── crf_refine.py
    ├── modelsize_estimate.py
    ├── test.py
    ├── train.py

Training and Testing

Please Note : The input images folder is always named 'train_A' and the GT folder is always named 'train_B' for uniform processing.

Training command :

python train.py

Testing command : The trained model for SDDNet can be download here: [Baidu Netdisk Link], code: mvpl or [Google Drive Link].

python test.py
python crf_refine.py
<!-- ## Evaluation We implement the widely-used metric, balanced error rate (BER). -->

Results

  1. Qualitative results: we provide the saliency maps, you can download them from [Baidu Netdisk Link], code: mvpl or [Google Drive Link].
  2. Quantitative results:

image

<!-- ## Bibtex ``` @article{HybridSOD, title={A weakly supervised learning framework for salient object detection via hybrid labels}, author={Cong, Runmin and Qin, Qi and Zhang, Chen and Jiang, Qiuping and Wang, Shiqi and Zhao, Yao and Kwong, Sam }, journal={IEEE Trans. Circuits Syst. Video Technol. }, year={early access, doi: 10.1109/TCSVT.2022.3205182}, publisher={IEEE} } ``` -->

Contact Us

If you have any questions, please contact Runmin Cong at rmcong@sdu.edu.cn or Yuchen Guan at yuchenguan@bjtu.edu.cn.