High-efficiency light field microscopy reconstruction for VCD-LFM. todo: add reference
- /datapre: Matlab scripts (GUI interface provided) for VCD-Net training dataset synthesis
- /vcdnet: Source code for deep-learning based light field microscopy reconstruction
- Python 3
- (Optional but highly recommended) CUDA 10.2 and CUDNN
- Packages:
- easydict==1.9
- imageio==2.4.1
- numpy==1.15.4
- scikit-image==0.14.1
- scipy==1.2.0
- tensorflow-gpu==1.14.0
- Matlab
For usage demo, we created a jupyter notebook to navigate through the pipeline. Example data can be downloaded from here. After unzip, put the data/
and checkpoint/
folder under vcdnet/
.
In current revision, we added an option to switch to a modified model for function imaging data reconstruction.
To check the model, go to /vcdnet/model/unet.py->UNet_B
The general pipeline stays the same and the only thing to do for model switching is to change from
config.model = 'structure'
to
config.model = 'function'
in /vcdnet/config.py
Edit the /vcdnet/config.py to ensure:
config.PSF.n_slices = 31
label = 'neuron_8um_simu_40x_n11_[m30-0]_step1um'
config.model = 'function'
config.VALID.lf2d_path = './data/to_predict2/'
And run
cd /code/vcdnet
python3 eval.py
or
cd /code/vcdnet
python3 eval.py --cpu
to reconstruct the demo images.
More examples were added to the example dataset.
Edit the /vcdnet/config.py to ensure:
config.PSF.n_slices = 61
label = 'beads_empirical_40x_n11_[m30-30]_step1um'
config.model = 'structure'
config.VALID.lf2d_path = './data/to_predict_beads/'
And run
cd /code/vcdnet
python3 eval.py
Edit the /vcdnet/config.py to ensure:
config.PSF.n_slices = 31
label = 'worm_empirical_40x_n11_[m30-0]_step1um'
config.model = 'function'
config.VALID.lf2d_path = './data/to_predict_worm/'
And run
cd /code/vcdnet
python3 eval.py