NVIDIAGameWorks / kaolin-wisp

NVIDIA Kaolin Wisp is a PyTorch library powered by NVIDIA Kaolin Core to work with neural fields (including NeRFs, NGLOD, instant-ngp and VQAD).
Other
1.45k stars 133 forks source link

Updated grid + AS api #85

Closed orperel closed 1 year ago

orperel commented 1 year ago

Originally, Triplanar grid used a single level-octree as an acceleration structure, essentially an AABB. Since we treated that accelstruct as an octree everywhere that caused some edge cases issues with the gui (i.e. triplane wouldn't draw any layers at all). In addition - the datalayers erroneously operated on BLASGrids and not the acceleration structs (the version we have right now draws information about occupancy).

This MR:

  1. Fixes errors in the design of the datalayers
  2. Fixes the datalayer issues that Triplane had before (makes sure AABB has drawable layers)
  3. Fixes how blas-grid / accelstruct names are passed everywhere, this is now more general (the previous if-else way was brittle)