This project aims to segment cell images of the GFP-GOWT1 dataset from the cell tracking challenge.
The module blobs models cells as ellipses, more specifically as gaussian functions, parameterized by their center, a matrix and a brightness.
The module stage1 models cells as polygons, more specifically as triangle meshes, that are rendered with the help of the pyredner raytracer. The results from the blob module can optionally be used as a starting point of this, more fine grained, optimization.
The module stage2 takes the produced vertex lists from stage1, and creates a dataset of cutouts from the original images. With the help of the ground truth segmentation masks these cutouts are either labeled as containing or not containing an actual cell.
The module train has a number of functions to train different classifiers on the dataset generated by stage2.
The module validate, with the help of train, first trains different classifiers on part of the data, then evaluates their performance on a separate dataset.
The module setup defines various hyperparameters, that are used throughout different modules. It also contains miscellaneous helper functions.
The other modules contain helper functions, for dealing with cells, setting up a scene with pyredner, loading and visualizing data, etc.
Requirements are
- pytorch
- numpy
- matplotlib
- sortedcontainers
- scikit-image
- pyredner and redner