3D Gaussian Adaptive Reconstruction for Fourier Light-Field Microscopy

Chenyu Xu^†, Zhouyu Jin^†, Chengkang Shen, Hao Zhu, Zhan Ma, Bo Xiong^*, You Zhou^*, Xun Cao, Ning Gu

^†Indicates Equal Contribution
^*Corresponding Author

Code repository of 3D Gaussian Adaptive Reconstruction for Fourier Light-Field Microscopy.

System Requirements

• Ubuntu 22.4
• cuda 12.7

Python Packages Requirements

conda env create -f environment.yml
conda activate 3dgat
pip install -e submodules/us_gaussian_voxelization
pip install -e submodules/simple-knn

Data Preparation

We provide the reticular data displayed in our main text (Fig. 2d) and supplementary material (Fig. S5) as examples. They are located in data/flfm/reticular1 (Fig. 2d) and data/flfm/reticular2 (Fig. S5) so you can test them easily.

For your own data, it should be located in data/<your_data_name> and the directory should contain:

• data.tif: ground truth volume, only exists in simulation experiments, with shape of [D, H, W]
• H.mat: the entire system PSF, with shape of [D, M, N], storded with sparse format
• psf.mat: 4D PSF cropped from the entire system PSF, according to the MLA center indexs in image coordinates, with shape of [V, D, Y, X]
• LFimage.tif: raw FLFM image, with shape of [M, N]
• MLcenter_idxes.mat: MLA center indexs in image coordinates
• params.yaml: data-related arguments for 3DGAT, typically the volume size to be reconstructed, and the image size of raw FLFM image
• deconvolve.yaml: arguments for wiener filtering and RL deconvolution

Quick Start

There are 2 steps for a quick start to perform our 3DGAT pipeline on the example data.

1. Wiener Filtering

Wiener filtering is firstly performed to get an initial estimation.

python onestep_wiener.py --root_dir data/flfm/reticular1

The Wiener filtering results will be stored in the specified root_dir

2. 3D Gaussian Adaptive Reconstruction

Then the Wiener filtering result is used as the initialization for 3D Gaussian Adaptive Reconstruction. As an example, we provide a reconstruction script for the example data (Fig. 2d in our main text, scripts/reticular-3dgat.sh). Some necessary arguments are also given in this scripts. More detailed description of these args can be found in opt.py.

bash scripts/reticular-3dgat.sh

Reconstruction results will be stored in results/GS/reticular

Comparison

We also provide the other two methods of RL deconvolution and gradient-based optimization baseline for comparison, as we discussed in our paper.

For RL deconvolution,

python flfm_RLdeconv.py --root_dir data/flfm/reticular1

For gradient-based optimization baseline, the reconstruction script for the example data (Fig. 2d in our main text) is also provided in scripts/reticular-baseline.sh.

bash scripts/reticular-baseline.sh

Citation

If you find our paper helpful, please cite us

@article{xu2025-3dgat,
  title={3D Gaussian adaptive reconstruction for Fourier light-field microscopy},
  author={Xu, Chenyu and Jin, Zhouyu and Shen, Chengkang and Zhu, Hao and Ma, Zhan and Xiong, Bo and Zhou, You and Cao, Xun and Gua, Ning},
  journal={Advanced Imaging},
  volume={55001},
  pages={1},
  year={2025}
}