Unsupervised Deep Unrolling Networks for Phase Unwrapping

Here we provide the official implementation of the CVPR-24 paper, unsupervised deep unrolling networks for phase unwrapping.

Information

• Authors: Zhile Chen (cs_zhilechen@mail.scut.edu.cn); Yuhui Quan (csyhquan@scut.edu.cn); Hui Ji (matjh@nus.edu.sg)
• Institutes: School of Computer Science and Engineering, South China University of Technology; Department of Mathematics, National University of Singapore
• For any question, please send to cs_zhilechen@mail.scut.edu.cn
• For more information, please refer to: [website]

Requirements

Here lists the essential packages needed to run the script:

• python 3.7.15
• pytorch 1.8.1
• numpy 1.21.6

Start Training

1. Download the training and test datasets provided in the Google Drive. Place them under the directory './data', e.g., './data/MoGR training data.hdf5'.
2. Run the training script, e.g.,

python train.py --lr 1e-3 --batch_size 10 --stage_num 3 --start_epoch 0 --distil_epoch 200 --end_epoch 700 --scheduler 'exp' --gamma 0.99 --expe_name 'PU_MoGR_Train' --traindata_id 'MoGR training data'

Test

Directly run the test script, e.g.,

python test.py --batch_size 10 --stage_num 3 --model_id 'PU_MoGR_Train/params_dict_epoch700.pth' --testdata_id 'MoGR test data_10dB' --save

Citation

@inproceedings{chen2024unsupervised,
  title={Unsupervised Deep Unrolling Networks for Phase Unwrapping},
  author={Chen, Zhile and Quan, Yuhui and Ji, Hui},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={25182--25192},
  year={2024}
}