kreshuklab/hylfm-net

HyLFM-Net

Requirements

• CUDA-enabled GPU
• Linux OS (tested on CentOs Linux 7 (Core))
• [mini]conda (tested with miniconda 4.8.4)

Installation

Clone repository

Clone and navigate to this repository

git clone git@github.com:kreshuklab/hylfm-net.git
cd hylfm-net

Install hylfm conda environment

conda env create -f environment.yml
conda activate hylfm
conda develop .

Install time

The install time greatly depends on download speed (several hundred MB).
🕐 Without download (or very fast download), the installation takes around 9 min.

Demo

Activate hylfm conda environment

conda activate hylfm

[optional] Choose a CUDA device

A cuda device may be selected before running hylfm (default 0), e.g.

export CUDA_VISIBLE_DEVICES=3

Use Weights and Biases logging

• Login/Sign up at https://wandb.ai/login
• Get your api token at https://wandb.ai/authorize (you'll be ask to provide this token on the first hylfm run, specifically on import wandb)

Train HyLFM-Net on beads

python scripts/train_presets/beads.py

🕐 Excluding download time, this training configuration runs for approximately 6 hours (on a GTX 2080 Ti). Note that the network will likely not have fully converged; increase max_epochs to allow for longer training (HyLFM-Net beads used in the paper was trained for 26.5 hours).

Test HyLFM-Net on beads (no previous training required)

To download and test HyLFM-net beads run

python hylfm/tst.py small_beads_demo

🕐 Excluding download time, this test configuration runs for approximately 6,5 min in total with 12 s per sample (on a GTX 2080 Ti). Most time is spend on computing metrics.

Data used in paper

Wagner, N., Beuttenmueller, F., et al. Deep learning-enhanced light-field imaging with continuous validation. Nat Methods 18, 557–563 (2021).

https://doi.org/10.1038/s41592-021-01136-0

beads: https://www.ebi.ac.uk/biostudies/studies/S-BSST622

Medaka heart: https://www.ebi.ac.uk/biostudies/studies/S-BSST604

zebrafish neural activity: https://www.ebi.ac.uk/biostudies/studies/S-BSST633

On Your Data

• Implement a get_tensor_info function in hylfm/datasets/local/<your dataset group>.py analogously to hylfm/datasets/local/example.py.
• Add your DatasetChoice (defined in hylfm_types.py) and extent get_dataset_sections (in datasets/named.py) and get_transforms_pipeline (in datasets/transform_pipelines.py) analogously to DatasetChoice.beads_highc_a
• Train or test HyLFM-Net as described in Demo.

Settings

To overwrite default settings, like the number of worker threads per pytorch Dataloader, adapt hylfm/_settings/local.py (copy from hylfm/_settings/local.template.py)