training_data/README.md

cellfinder training data

This repository contains the data used to train the 3D ResNet-50 supplied with cellfinder, as described in Tyson, Rousseau & Niedworok et al. (2021).

Briefly, it contains ~100,000 sets cell candidate images, ~50,000 cells and ~50,000 non-cells. Each cell candidate (cell-like object) is centered in a three channel, 3D image (50x50x100um).

The data was acquired using serial section two-photon microscopy, as per Velez-Fort et al. (2018).