The HGCal graph NN-based reconstruction has not been officially introduced in CMSSW. Implementation in a SonicTriton producer is likely the easiest path (especially for a PyTorch graph-based network, since the PyTorch C++ API is not yet available in CMSSW). There are multiple networks to implement, including edge classification (for clustering) and dynamic reduction (for energy regression etc.).
Particular considerations:
Consider introducing ProcessModifier to enable the alternative sequence
kpedro88
commented
3 years ago
The HGCal graph NN-based reconstruction has not been officially introduced in CMSSW. Implementation in a SonicTriton producer is likely the easiest path (especially for a PyTorch graph-based network, since the PyTorch C++ API is not yet available in CMSSW). There are multiple networks to implement, including edge classification (for clustering) and dynamic reduction (for energy regression etc.).
Particular considerations:
Assigned to: @tklijnsma, @lgray