List view

• Decide array implementation

  We want an array library that supports both CPU and GPU with sufficiently rich set of operations.

  Overdue by 9 month(s)

  •

  Due by April 1, 2025

  •1/1 issues closed
  100% complete0 open1 closed

• Demonstrate consistent ROI results between SPNG and OSP

  This milestone is similar to milestone/2 but can be independent. It may use validated SPNG decon or OSP decon as input.

  No due date
  0% complete0 open0 closed

• Define representation for ROI finding

  This requires defining conventions for Tensor representations of decon waveform (with different filterings) as well as ROIs. ROI representation might be some form of masked tensor or a small type int (bool, int8) giving 0/1 values.

  No due date
  0% complete0 open0 closed

• Port OSP "conventional" ROI finding to SPNG

  This in addition to "simply" using Torch tensors, this requires inventing new GPU/SIMD-friendly algorithms equivalent to the various heuristic algorithms in OSP. The SPNG ROI finding should be factored into a number of different WCT DFP graph nodes. Some of these will operate at the data scope of a single wire plane, others will require bringing together intermediate ROI representation across planes.

  No due date
  0% complete0 open0 closed

• Ability to run SPNG deconvolution in a subgraph consuming and producing IFrame

  Enough SPNG developed to allow a WCT DFP graph ``` IFrame -> [SPNG subgraph] -> IFrame ```

  No due date
  0% complete0 open0 closed

• Demonstrate consistent decon results between SPNG and OSP

  Using common input, which may be simulation, but which spans tracks with multiple directions, we demonstrate nearly identical results from the deconvolution between SPNG and OSP. To perform this, we will use OSP's feature to extract the "decon" tagged traces.

  No due date
  0% complete0 open0 closed

• Move spng to become wire-cell-toolkit subproject

  No due date
  0% complete0 open0 closed