/ImFCS_FCSNet_ImFCSNet

Training code for FCSNet and ImFCSNet

Primary LanguagePythonMIT LicenseMIT

Deep learning reduces data requirements and allows real-time measurements in Imaging Fluorescence Correlation Spectroscopy

This repository contains the training code of FCSNet and ImFCSNet models.

Create the conda environment

Create a conda environment called tf25_gpu with the environment.yml file. The environment contains TensorFlow v2.5.

conda env create -f environment.yml

Training FCSNet

To train FCSNet on a NVIDIA GPU, navigate to fcsnet folder and run the main.py file.

cd fcsnet
conda activate tf25_gpu
CUDA_VISIBLE_DEVICES=0 python main.py

The FCSNet training data can be placed in training_data/data_EMCCD and training_data/data_Gaussian folders. They are the simulated 2D/3D ACF curves with the respective noise type. You can change the seed and the hyperparameters of the network in CNN.py file to try different experiments.

Training ImFCSNet

Similarly, navigate to the imfcsnet folder and run the main.py file.

cd imfcsnet
conda activate tf25_gpu
CUDA_VISIBLE_DEVICES=0 python main.py

You can amend the configuration files, which are in the Configurations folder, to make changes to the ImFCSNet training. You can make the following changes to try different experiments.

  1. In CNN.py file:

    • 2D/3D training: IS_2D variable
    • seed: GLOBALSEED variable
    • re-train a model: IS_RETRAINING and retrain_model_fn variables
    • learning rates: INITIAL_LEARNING_RATE and FINAL_LEARNING_RATE variables
    • configuration of the network

  2. In fd1t_cfg_2d.py or fd1t_cfg_3d.py files, you can change the range of simulation 2D/3D parameters.

License

Distributed under the MIT license. See LICENSE for details.