microsim
Light microscopy simulation in python.
The goal of this library is to generate highly realistic simulated data such as the following:
(this data was generated using segmentations from
cell jrc_hela-3 at OpenOrganelle as the ground truth)
https://github.com/user-attachments/assets/905fa1af-c8bd-406e-87a7-7c5921de74d1
Documentation
Documentation is at https://talleylambert.com/microsim/. Start with
First simulation,
read the User guide for how each part
of the simulation works, and see the
API reference for every option.
[!WARNING]
microsim is in early development and changes quickly. Releases may break the API
or change simulation results for the same parameters. If you depend on it, pin a
specific version and record it with your results.
Installation
from PyPI
pip install "microsim[all]"
[!NOTE]
At the moment, this library is a bit more "application" than it
is "library". If you are following the docs or tutorials, its
probably best to install the full [all] extra, which brings in
io, visualization, and other data-fetching dependencies. However,
the bare minimal install is fully functional for generating data.
Local Development
gh repo clone tlambert03/microsim
cd microsim
uv sync
From github
To get the bleeding edge version, which will change rapidly, you can install from github.
pip install "microsim[all] @ git+https://github.com/tlambert03/microsim"
With GPU support
If available, microsim can use either Jax or Cupy to accelerate computations.
These are not installed by default, see the
jax
or cupy installation instructions,
paying attention to your GPU requirements. Support for torch is planned.
Usage
Construct and run a
microsim.Simulation
object.
from microsim import schema as ms
from microsim.util import ortho_plot
# define the parameters of the simulation
sim = ms.Simulation(
truth_space=ms.ShapeScaleSpace(shape=(128, 512, 512), scale=(0.02, 0.01, 0.01)),
output_space={"downscale": 8},
sample=ms.Sample(labels=[ms.MatsLines(density=0.5, length=30, azimuth=5, max_r=1)]),
modality=ms.Confocal(pinhole_au=1),
output_path="au1.tiff",
)
# run it
result = sim.run()
# optionally plot the result
ortho_plot(result)
