Home

Awesome

LSGM

Paper: Lens-less imaging via score-based generative model
https://www.opticsjournal.net/M/Articles/OJf1842c2819a4fa2e/Abstract

Authors: Chunhua Wu, Hong Peng, Qiegen Liu, Wenbo Wan, Yuhao Wang

Date : October-31-2022
Version : 1.0
The code and the algorithm are for non-comercial use only.
Copyright 2022, Department of Electronic Information Engineering, Nanchang University.

Lens-less imaging is affected by twinning noise in in-line holograms, and the reconstructed results has always faced the poor reconstruction signal-to-noise ratio and low imaging resolution. In this paper, a lens-less imaging via score-based generation model is proposed. In the training phase, the proposed model perturbs the data distribution by slowly adding Gaussian noise using a continuous stochastic differential equation (SDE). Acontinuous-time dependent score-based function with denoising score matching is then trained and used to solve the inverse SDE to generate object sample data. In the test phase, a single Fresnel ZoneAperture is used as a mask to achieve lens-lessen coding modulation under incoherent illumination.The prediction-correction method is then used to alternate iterations between the numerical SDE solver and data-fidelity term steps to achieve lens-less imaging reconstruction. Validation results on LSUN-bedroom and LSUN-church datasets show that the proposed algorithm can effectively eliminate twin image noise, andthepeaksignal-to-noiseratioand structural similarity ofthereconstructionresultscanreach 25.23dBand 0.65.The PSNR values of the reconstruction results are 17.49 dB and 7.16 dB higher than lens-less imaging algorithms based on traditional Back Propagation or Compressed Sensing, respectively. The corresponding SSIM values were 0.42 and 0.35 higher, respectively. Thereby,the reconstruction quality of the lensless imaging is effectively improved.

Requirements and Dependencies

python==3.7.11
Pytorch==1.7.0
tensorflow==2.4.0
torchvision==0.8.0
tensorboard==2.7.0
scipy==1.7.3
numpy==1.19.5
ninja==1.10.2
matplotlib==3.5.1
jax==0.2.26

Checkpoints

We provide pretrained checkpoints. You can download pretrained models from [Baidu cloud] (https://pan.baidu.com/s/12SzKZRNefjNqOx_nW-2RAA))

Dataset

The dataset used to train the model in this experiment is LSUN-bedroom and LSUN-church.

place the dataset in the train file under the church folder.

Train:

python main.py --config=configs/ve/church_ncsnpp_continuous.py --workdir=exp_train_church_max1_N1000 --mode=train --eval_folder=result

Test:

python score_sde_fza_demo_fujian.py

Flow chart of lens-less imaging

<div align="center"><img src="https://github.com/yqx7150/LSGM/blob/main/fig1.png"> </div>

Training and reconstruction flow chart of LSGM algorithm

<div align="center"><img src="https://github.com/yqx7150/LSGM/blob/main/fig4.png"> </div>

Results on experiment data

<div align="center"><img src="https://github.com/yqx7150/LSGM/blob/main/fig5.png"> </div>

Visual comparison of reconstruction images on the LSUN-bedroom dataset.

Acknowledgement

The implementation is based on this repository: https://github.com/yang-song/score_sde_pytorch.

Other Related Projects