Closed twongjirad closed 8 years ago
Added in https://github.com/LArbys/LArCV/pull/74.
For training purpose, this is already possible with SegFiller on the fly w/o making a new file. When we train single particles, we combine Pi0=>Gamma, Kaon=>Pion, etc. Another thing SegFiller is useful for is that it can re-label on the fly w/ 0-indexing labels instead of using enum values, so it is a natural translation to how it's used by caffe for instance. Training for MIP/HIP/Shower segmentation is tried out using SegFiller indeed!
For the near future, we might want to relabel semantic segmentation label images so that individual particles can be relabeled with more inclusive categories.
Specific example is to provide the same label for "MIP tracks" which include muon and charged pions. Similarly, label "shower tracks" for electrons, gammas, and pi-zeros.
Would be good to have a model that can redo this labeling.