search

AFD-Illinois / ipole

Polarized covariant radiative transport code
16 stars 12 forks source link

Preliminary adaptive resolution #26

Closed bprather closed 4 years ago

bprather commented 4 years ago
bprather commented 4 years ago

This is currently blocked on:

  1. Slow light test, to make sure I haven't broken it
  2. Adaptive res regression test, to ensure adaptive images never look truly crazy compared to full-res
  3. Knowing what to do w.r.t. interpolation. Currently uses straight bi-linear, but interpolating from interpolated pixels is fuzzy. May need to keep track of calculated vs interpolated pixels & use just the former. The heuristic could always use work, suggestions welcome...
bprather commented 4 years ago

This branch has progressed far enough to be merged I think. Quick summary:

  1. New heuristic from Zack Gelles
  2. Regression test that a suitably stringent refinement requirement produces an image with very low MSE. Note that compare.py return criterion is now MSE, it was relative flux difference for some reason.
  3. While there's still work in figuring out interpolation, the code already tracks which pixels were traced vs interpolated, so it is unlikely to need major changes In short this solves blocker (2) and lays a good foundation for (3). (1) will take longer and has its own bug, #24