🔒 Locked 🎮 Unity

Why transform Normal par WorldViewInvertTranspose in Nvidia Cg Shader

Started by ClementLuminy Dec 29, 2006 at 12:10 PM 3 replies 5.7k views

Original Post

Dec 29, 2006 at 12:10 PM

Hello. I'm wondering why we have to multiply the Normal of a vertex ( expressed in objectd space ) by the "invert transpose World view matrix" ... Most of the sample which are using NVidia Cg language to do very simple lighting ( per-vertex directional lighting ) use the following pseudo code in order to transform input data to the correct space: =========== float4 WorldPos = mul( Input.Pos, WorldMatrix ); float3 WorldNormal = mul( Input.Normal, (float3x3) WorldViewIT ); =========== For me, with my very poor mathematics skill :) this pseudo code is mode "logic": =========== float4 WorldPos = mul( Input.Pos, WorldMatrix ); float3 WorldNormal = mul( Input.Normal, (float3x3)WorldMatrix ); =========== you transform the normal the same way you the position ( execpt that it's a vector and not a position ..) So, what's the mathematic tricks behind this normal transformation ??? Why do i have to use the WorldViewIT ( And Why Invert Transpose moreover !!!! ) instead of the simple world Matrix? Thanks a lot for your help !

_swx_

Dec 29, 2006 at 9:50 PM

http://www.cs.utah.edu/classes/cs6610/lectures/MathematicsOfPerPixelLighting.pdf has a good explanation

wyrzy

Dec 29, 2006 at 11:31 PM

The above link does a decent explanation, but, an alternate (algebraic) way of thinking about the problem is that is v . n == 0 is Model Space (where n is the normal and v is the vertex), then v' . n' should equal 0 in View Space.

Let V be the matrix that transforms verticies and
N be the matrix that transforms normals, then

v' . n' == 0
=> (v')T * n' == 0 (where (v')T denotes v' transposed and * is matrix mul)
=> (V(v))T * N(n) == 0 (substitution)
=> (vT * VT) * N(n) == 0 (expand the transpose operator)
=> vT * (VT * N) * n == 0 (matrix multiplication is associative)

since vT * n == 0, then
VT * N == I
=> VT^(-1) * VT * N == VT^(-1)
=> N == VT^(-1) (the inverse transpose of the model view matrix)

Now, it should also be noted that if your model view matrix only contains rigid body rotations (generally translations and rotations), then VT^(-1) == V (because the transpose of a rigid body transformation is its inverse).

grimreaper85

Dec 30, 2006 at 12:20 AM

So the inverse transpose is only needed if you have scaling in your WorldMatrix ?

otherwise you can just use (float3x3)WorldMatrix ?

ClementLuminy

Dec 30, 2006 at 6:46 AM

Hello !

Thanks for your help.

The things is that in the NVidia document listed above, every computation happen in eye space...

This seams to me pretty strange because we have to transform any other information expresed in wolrd space into eye space ( i.e: light direction .... )

So why do they do all their computations in eye space ? why don't they do them in World space ? this solution seams to me much more simpler, but i think i missunderstood something...

Another things about the mathematical explanation ....
can you re explain it one time, but in a more detailed way ..?..?

Thanks again!

Clement

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.