castano

                 RMSE           PSNR            Time

Kodak/Waterloo Image Set:

stb              8.202766       29.851597       0.258041
stb-hq           8.009301       30.058910       0.284019
nvtt-fast        8.089954       29.971882       0.445670

nvtt             7.616215       30.496019       6.806233
nvtt-hq          7.562366       30.557650       13.081200

castano

/*===========================================================================
	Title: BSpline.cpp
	Module: Pi/MathLib
	Author: Ignacio Castaño
	Date: 29/05/2000
	License: Public Domain
===========================================================================*/

/*----------------------------------------------------------------------------
	Doc:

castano

I didn't think there was much to gain from optimizing quantization intervals for vertex compression, but I thought it would give it a try and see it for myself.

I have a set of vertices X that I want to quantize. The trivial way to do that is to transform them to the [0,1] range and store them using an integer UNORM format.

At runtime the vertices are reconstructed as follows:

X = Q * m + a

Where: