Phase unwrapping for fringe projection profilometry (FPP) using deep learning.
This code implements the approach as
described in the following research paper:
- Deep absolute phase recovery from single-frequency phase map for handheld 3D measurement
- Songlin Bai, Xiaolong Luo, Kun Xiao, Chunqian Tan and Wanzhong Song*
- Optics Communications, 2022(512) [PDF]
- Absolute fringe-order is retrieved from one FPP phase map by the DCNN
- The DCNN is lightweight and operates in real-time for a phase-map of 1024ร1024 pixels on a GTX 1660Ti.
- A large-scale and challenging phase unwrapping dataset is built from real objects and publicly available.
This code was developed and tested with python 3.6, Pytorch 1.8.0, and CUDA 10.2 on Ubuntu 18.04. It is based on Eduardo Romera's ERFNet implementation (PyTorch Version).
install manually the following packages :
torch
PIL
numpy
argparse
Our raw data SCU-Phase-RawData will be available.
Our ready dataset is SCU-Phase-ReadyData.
Training the HiPhase model from scratch on SCU-Phase-ReadyData by running
python train/main.py
Evaluating the trained model by running
python eval/eval_gray.py
Evaluating the mIoU by running
python eval/eval_iou.py
Our pretrained HiPhase model is HiPhase-experi
@article{Bai2022,
author = {Bai, Songlin and Luo, Xiaolong and Xiao, Kun and Tan, Chunqian and Song, Wanzhong},
title = {Deep absolute phase recovery from single-frequency phase map for handheld 3D measurement},
journal = {Optics Communications},
publisher = {Elsevier Ltd.},
volume = {512},
year = {2022}
}
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License, which allows for personal and research use only. For a commercial license please contact the authors. You can view a license summary here: http://creativecommons.org/licenses/by-nc/4.0/