Skip to content

Latest commit

 

History

History
93 lines (73 loc) · 3.17 KB

README.md

File metadata and controls

93 lines (73 loc) · 3.17 KB
Raw

3DM: Deep decomposition and deconvolution microscopy

Demo video acquired using 3DM.

Compression is applied, portion of video is shown due to the file size. Supplementary video 2 is the video before compression.

Paper

Official source codes for "3DM: Deep decomposition and deconvolution microscopy", Optics Express.

ABSTRACT

Requirements

python==3.7.6
torch==1.5.0
skimage==0.16.2

Demo: BEAR + Deconvolution with pretrained weight

For more detail about BEAR which does unsupervised low-rank and sparse decomposition, please refer to our previous paper. [PAPER][Code and dataset]

There are videos saved in this notebook, where they are not visible if you simply view the notebook in github. One way to view all things in the notebook is opening it with VS Code.

Executing shell (download + Notebook) takes approximately 10 minutes, but may differ to your network connection and computing power.

sh ./run_jupyter.sh

This notebook lets you:

  • Download a portion (t=1~50, size=1.5GB) of calcium imaging data acquired with our wide-field microscope. Note that the number of timeframes of whole data we used was 1260 with 4.2VPS.
  • Do unsupervised low rank and sparse decomposition using BEAR.
  • Load the pretrained 3-D deconvolution network.
  • Do deconvolution for each 50 sparse volumes.
  • Visualize the results.

Followings are visualization results in the notebook.

Unsupervised low rank and sparse decomposition using BEAR.

Maximum intensity projection of videos.

3-D Deconvolution using pretrained network.

Maximum intensity projection of videos. Finally, from raw video acquired using our wide-field microscope, we acquired sparse, deconvolved video!

Train

The following command starts training the 3-D deconvolution network:

python codes/train.py

Test

The following commands do deconvolution after loading the pretrained weight.

python codes/eval_simulation.py --exp_name 3DM --epoch 26000
python codes/eval_3DM_video.py --exp_name 3DM --epoch 26000
  • eval_simulation.py do deconvolution for simulated wide-field data. (See Section 3.1)
  • eval_3DM_video.py do deconvolution for wide-field data acquired with our microscope. (See Section 3.2)

Citation

@article{cho20213dm,
author = {Eun-Seo Cho and Seungjae Han and Kang-Han Lee and Cheol-Hee Kim and Young-Gyu Yoon},
journal = {Opt. Express},
number = {20},
pages = {32700--32711},
publisher = {OSA},
title = {3DM: deep decomposition and deconvolution microscopy for rapid neural activity imaging},
volume = {29},
month = {Sep},
year = {2021},
url = {http://www.osapublishing.org/oe/abstract.cfm?URI=oe-29-20-32700},
doi = {10.1364/OE.439619},
}