Random Forests for Multiclass Segmentation using Python API in PerGeos
Simple demonstration of feature computation where the feature vector contains intensity and 2D eigenvalues (X & Y). This example uses random forests implementation from the sklearn package.
This
Xtra includes a PerGeos project and the python script named ‘RandomForests_PerGeos_tutorial__all_slices_advanced.pyscroï‘
and as the name suggests, the script is part of a tutorial to showcase how to
use powerful machine learning tools using standard python libraries to achieve
multiclass segmentation. Feature
descriptor computation includes only two different features, (a) Grayscale
Intensity, (b) 2D EigenValues. User can provide sparse labels in order to
classify the rest of the volume. Inputs expected are (a) Grayscale volume/image
(b) Equal sized (as Grayscale volume) label volume with labels on the 0th
slice. This can be extended by choosing any voxels across other slices for the
sparse labels but the script currently requires labels only on the 0th
slice. Note that while the example project is for PerGeos, this script is
compatible for use with Amira-Avizo as well.