Skip to content

Releases: LandesLinkLab/6D-microscopy

Release list

v1.1 — version accompanying the published article

v1.1 — version accompanying the published article Latest
Latest

Choose a tag to compare

View all tags
@engillett engillett released this 01 Aug 18:48
v1.1
e14e4ea

This is the version of the code that accompanies the published article.

Article
Gillett, E., Chatterjee, S., Chatterjee, J., Kovalenko, N., Xu, C., Fan, D., Chen, Y., Qiu, Y., Miao, J., Nelavoy, V., Lew, M. D., Backlund, M. P., and Landes, C. F. "Fused deep learning enables 6D single-molecule localization in polarization-resolved microscopy." Methods and Applications in Fluorescence (2026).
https://doi.org/10.1088/2050-6120/ae8eff

Data
The processed data behind each figure, together with a snapshot of this code, are deposited in the Illinois Data Bank:
https://doi.org/10.13012/B2IDB-6080564_V1

Changes since v1.0

  • README now cites the published article DOI and the Illinois Data Bank dataset DOI, and adds a Data availability section.
  • Adds Unified_Preprocessing_Pipeline_v6 (end-to-end colocalization and cropping) and removes four large sample-run artifacts that were not pipeline code.

Large data files
The four .mat files for 4_mDeepSTORM3D/ (Din.mat, Dtar.mat, target_bol.mat, pred_bol.mat, 639 MB total) remain attached to the v1.0 release, since they are unchanged. Download them from there and place them in 4_mDeepSTORM3D/, or regenerate them with demo1.py and demo3.py.

Assets 2
Loading

v1.0 — mDeepSTORM3D training and demo data

v1.0 — mDeepSTORM3D training and demo data

Choose a tag to compare

View all tags
@engillett engillett released this 30 May 07:05
v1.0
89b3427

What's in this release

This v1.0 release attaches the four large MATLAB data files referenced by the 4_mDeepSTORM3D/ subfolder of the repository. They are too large for regular git push (GitHub's 100 MB per-file hard limit) and are hosted here as release assets instead.

File Size Purpose
Din.mat 168 MB DeepSTORM3D training input — simulated PSF image stack used to train the localization network. Produced by running 4_mDeepSTORM3D/DeepSTORM3D/GeneratingTrainingExamples.py (or demo1.py).
Dtar.mat 123 MB DeepSTORM3D training target — ground-truth localization labels paired with Din.mat. Produced by the same script.
target_bol.mat 176 MB Demo evaluation ground-truth volume. Produced by demo3.py against a pretrained model.
pred_bol.mat 176 MB Demo predicted output volume from the trained network. Produced by demo3.py.

Total: 641 MB.

How to use

  1. Download the four .mat files from the Assets section below.
  2. Place them inside 4_mDeepSTORM3D/ in your local clone of this repository, next to demo1.py...demo5.py.
  3. The demos and the training scripts expect them at that path.

Alternatively, you can regenerate all four files from scratch by running the demos — this is how the original DeepSTORM3D (Nehme et al. 2020) intends users to obtain the data. Training takes ~30 hours on a Titan Xp; demo evaluation takes seconds.

Citation

If you use this code or data, please cite:

Gillett, E., Chatterjee, S., Chatterjee, J., Kovalenko, N., Xu, C., et al. Fused deep learning enables 6D single-molecule localization in polarization-resolved microscopy. Methods and Applications in Fluorescence (2026), MAF-101478.

License

See LICENSE (and LICENSE-Landes in 4_mDeepSTORM3D/) in the repository. Non-commercial educational and research use only.

Loading