Podtyazhki1337/3d-neurons-segmentation

A large-scale multi-species resource reveals cross-species generalization in label-free 3D neuron segmentation

The code is available here: https://github.com/podtyazhki1337/3d-neurons-segmentation-benchmark The dataset is available on Zenodo: https://zenodo.org/records/20797635 This repository contains trained model weights used in the study:

The models were trained for 3D neuron segmentation in label-free microscopy across:

• species: rat, mouse, human
• modalities: oblique illumination and Dodt gradient contrast (DGC / DODT)

Repository contents

The repository includes trained weights for five model families:

• nnU-Net v2
• 3D U-Net
• StarDist3D
• Cellpose
• micro-SAM

Weights are organized into separate folders by model family and training configuration.

This dataset, code and weights accompanies the manuscript: "A large-scale multi-species resource reveals cross-species generalization in label-free 3D neuron segmentation."

If you use this weights in your work, please cite the associated publication once available.