Function: createNTCGridPyramidModel()
createNTCGridPyramidModel(
options,random):NTCGridPyramidModel
Creates the CPU-side reference model: a genuine mip pyramid of feature
grids (resolutions/grids, finest-first - see computeGridLevels) plus
a small MLP decoder.
At any given LOD, exactly one grid level's channels-wide feature vector
(selected via NTCMipBands.selectFeatureLevel - not concatenated with any
other level's) feeds the decoder, alongside the normalized LOD itself -
which the decoder needs because a single stored level is reused to
reconstruct several different physical mips (see NTCMipBands.js's doc
comment). This is what makes the decoder's input a fixed channels + 1
wide regardless of how many mip levels the pyramid stores, and it mirrors
the NVIDIA neural texture compression paper's own decoder input (Section
4.4: one feature level's taps plus a LOD value).
Parameters
options
random
() => number