Skip to content

Code Notes for post processor code

twongjirad edited this page Aug 30, 2018 · 2 revisions

This is an attempt to supplement documentation within the code comments with the larger picture behind each of the components. This is meant to help new developers more efficiently parse the code by explaining the larger goal and what was intended. The hope is to help people understand how to make improvements.

This covers the larflow "postprocessor", the set of algorithms to help us take CNN outputs -- infill, endpoint, and larflow -- and produce 3D cluster reconstruction that we want to use to help reject cosmic ray muon tracks and identify in-time neutrino clusters.

Right now the code is in the early development phase. Trying to get things in quickly and get to the end to see what kind of base performance we are working with. Will also help us understand where to put first efforts for optimization.

Code lines in the postprocessor folder in the larflow repo.

Descriptions of code are as follows

  • dev.cc: program to run algorithm chain. geared to operate on fixed test-data sample for now. Go to testdata folder in the larflow repo for a script that will help you download the testdata inputs
  • ContourTools: centers around a 2D, single-plane algorithm that clusters charge into "straightest line segments" by using contour analysis, namely defect detection.
  • FlowContourMatch: uses larflow network input and countour clustering algo to define the 3D position for either pixels or hits on the Y plane.

Clone this wiki locally