LinkinPy-StarDist

linkinpy-stardist provides LinkinPy-compatible StarDist 2D training and prediction workflows. The package is intended to be installed through Library/StarDist.yaml, but it can also be used directly from Python.

Workflows

• train_2d_with_qc: train a custom StarDist 2D model with optional augmentation and held-out quality control.
• predict_2d_custom_model: run inference with a trained model directory.
• predict_2d_pretrained_model: run inference with a StarDist pretrained model.
• batch_predict_2d_custom_model: run custom-model prediction over an input folder.
• batch_predict_2d_pretrained_model: run pretrained prediction over an input folder.

Install

uv pip install linkinpy-stardist

Most users should install it indirectly:

cd LinkinPy
uv run linkinpy install ../Library/StarDist.yaml --interface cli

Data Layout For Training

Training expects paired raw images and instance-label masks. File names are matched by stem:

training/
  images/
    cell01.tif
    cell02.tif
  masks/
    cell01.tif
    cell02.tif

images/cell01.tif pairs with masks/cell01.tif.

CLI Example Through LinkinPy

cd LinkinPy
uv run linkinpy run ../Library/StarDist.yaml "Predict 2D Pretrained Model" \
  --output-dir outputs \
  image=input.tif \
  model_name=2D_versatile_fluo

Prediction methods return labels as np.ndarray outputs. Batch methods return summary paths that point to written batch results.

Direct Python Example

import tifffile
from linkinpy_stardist import predict_2d_pretrained_model

image = tifffile.imread("input.tif")
labels = predict_2d_pretrained_model(
    image=image,
    model_name="2D_versatile_fluo",
)

Quality Control

When quality control is enabled during training, the data is split into train, validation, and test sets. The held-out test split is used only for post-training QC metrics.

For the full LinkinPy interface tutorial, see ../../USAGE_TUTORIALS.md.