Datasets:

---

OpenScientificCodeRegistry

/

Database

Like 0

Follow

Open Scientific Code Registry 1

ERROR: type should be string, got "\nhttps://github.com/user-attachments/assets/1ad2210c-da95-45f6-ab59-27d5647b267a\n\n# LUNAR: Localization Using Neural-physics Adaptive Reconstruction\n\nThis repository contains the official implementation of the paper [\"Aberration-aware 3D localization microscopy via self-supervised neural-physics learning\"](https://www.nature.com/articles/s41467-026-73045-9), published in *Nature Communications* (2026).\n\nIf you find this repository useful for your research, please cite: \nFu, S., Shi, W., Katrukha, E.A. et al. *Aberration-aware 3D localization microscopy via self-supervised neural-physics learning*. *Nature Communications* (2026). https://doi.org/10.1038/s41467-026-73045-9\n\nA brief video introduction to the characteristics of LUNAR:\nhttps://github.com/user-attachments/assets/1ad2210c-da95-45f6-ab59-27d5647b267a\n\n\n* whole-cell Nup96 NPC imaging with a 6 μm DMO Tetrapod PSF\n<p align=\"left\">\n <img src=\"docs/NPCmovie1~2.gif\" width='500'>\n</p>\n\n* Neuron imaging at 50 μm depth in brain slice.\n<p align=\"left\">\n <img src=\"docs/Neuron_movie1~1.gif\" width='500'>\n</p>\n\n* 20 μm thick whole-cell reconstruction with motor-PAINT and lattice light-sheet microscopy.\n<p align=\"left\">\n <img src=\"docs/LLS-motor-PAINT~1.gif\" width='500'>\n</p>\n\n## Installation\nThe code was tested on Windows 11 and Ubuntu 22.04.5 LTS system, \nwith software environment managed by Anaconda.\nPackages required can be found in `requirements.txt`. \n\nTo install LUNAR, please follow the steps below:\n\n1. Clone the repository:\n```commandline\ngit clone https://github.com/Li-Lab-SUSTech/LUNAR.git\ncd LUNAR\n```\n\n2. Create a new conda environment:\n```commandline\nconda create --name ailoc python=3.9.7 \n```\n\n3. Activate the environment and install the required packages:\n```commandline\nconda activate ailoc\npip install -r requirements.txt\n```\n\n4. Run demos:\n```commandline\n# get into the demos directory\ncd demos\n\n# run the demos sequentially\npython demo1-simu_tubulin_tetra6.py\npython demo2.1-psf_calibration.py\npython demo2.2-exp_npc_dmo1.2.py\npython demo3-exp_whole_cell_tetra6.py\n```\n\nThe project files should be organized as the following hierarchy:\n```\n├── root\n│ ├── ailoc\n│ │ ├── common // common tools used by all algorithms\n│ │ │ ├── xxloc.py // the abstract class xxloc\n│ │ │ ├── vectorpsf_fit // module for vector PSF calibration from beads\n│ │ │ ├── analyzer.py // utilize xxloc to analyze experimental data\n│ │ │ ├── ... // other common modules, such as notebook_gui, utilities...\n│ │ ├── simulation // simulation tools\n│ │ │ ├── simulator.py // the simulator class for generating SMLM data\n│ │ │ ├── ... // other simulation modules such as PSF, camera, etc.\n│ │ ├── deeploc // the implementation of deeploc\n│ │ │ ├── deeploc.py // deeploc class, which inherits from xxloc\n│ │ │ ├── network.py // the network of deeploc, implement with forward function\n│ │ │ ├── loss.py // the loss function of deeploc\n│ │ ├── lunar // the implementation of LUNAR\n│ │ │ ├── lunar.py // lunar class, which inherits from xxloc\n│ │ │ ├── network.py // the network of lunar, implement with forward function\n│ │ │ ├── ... // other lunar modules, such as loss, submodules, etc.\n│ │ ├── ... // other xxloc algorithms, fd-deeploc, etc.\n│ ├── datasets // test dataset should be put here\n│ ├── results // results will be saved here by default\n│ ├── usages // general usage scipts for reference\n│ │ ├── notebook // notebook with GUI\n│ │ ├── pyscripts // python scripts\n│ ├── demos // several demos to show how to use the framework\n```\n\n## Demos\nWe recommend testing the code on a work station with an NVIDIA GPU with 24 GB memory, and RAM >= 64 GB.\nIf such a GPU is unavailable, \nyou can reduce the `batch_size` or `context_size` to lower GPU memory usage during training. \n(e.g., `batch_size=1`, `context_size=8`)\n\nWe provide several demos in the `demos` directory, illustrating use cases such as beads calibration, \nlocalization learning based on the calibration, and synchronized learning directly on raw data. \nFor all demos: \n\n* Data should be downloaded from [](https://doi.org/10.5281/zenodo.14709467)\n and put in the `datasets` directory.\n* Results will be automatically saved in the `results` directory.\n* Localizations are saved as a `.csv` file. \n\nFor post-processing and visualizing the localizations, \nWe recommend using the [SMAP](https://www.nature.com/articles/s41592-020-0938-1) software, \nwhich supports tasks like rendering, filtering, and drift correction, etc.\n\n### Demo 1: Simulated Microtubule Dataset\nDemo 1 is based on simulated microtubule datasets with 6 μm Tetrapod PSF. \nYou need to download the dataset.\nThe script `demo1-simu_tubulin_tetra6.py` trains DeepLoc and LUNAR LL with an accurate PSF, followed by LUNAR SL with a wrong PSF. \nAll models were trained once with fixed density and tested on three datasets with low, medium, and high densities, respectively. \nEvaluation metrics are automatically printed out after each model's training.\n\n### Demo 2: Experimental Nup96 NPC Dataset\nDemo 2 is based on the experimental Nup96 NPC dataset with a 1.2 μm DMO Saddle-Point PSF. \nThe dataset and beads for PSF calibration are provided. \nNote that this dataset has ~60 nm y-drift during data acquisition. \n\n* `demo2.1-psf_calibration.py`: Calibrate the PSF; calibration results are saved along with the beads file.\n* `demo2.2-exp_npc_dmo1.2.py`: First train DeepLoc with the calibrated PSF, followed by DeepLoc and LUNAR SL with a wrong astigmatic PSF.\n\n### Demo 3: Whole-Cell Nup96 NPC Dataset\nDemo 3 is based on the whole-cell Nup96 NPC dataset with a 6 μm DMO Tetrapod PSF. \nThe script `demo3-exp_whole_cell_tetra6.py` trains DeepLoc and LUNAR SL with a mismatched PSF.\n\n### Demo 4: Notebooks with GUI for Demo 2\nDemo 4 provides three Jupyter notebooks with graphical user interface (GUI) to demonstrate PSF calibration, model learning, and model inference. \nThis demo uses the same data as Demo 2. \nA video `docs/demo4_video_1~1` is provided to show the usage of these notebooks.\n"