Jump to content

  • Log In with Google      Sign In   
  • Create Account


normal mapping messing up cos angle ?


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
2 replies to this topic

#1 lipsryme   Members   -  Reputation: 878

Like
0Likes
Like

Posted 11 October 2012 - 09:11 AM

I'm having problems with the lighting in my shader when using normal mapping instead of just passing the normals.
Better look at the screenshots (angles are all correct without normal mapping, directional light source (0, 0, 1))

without normal mapping front:
http://cl.ly/image/051a3g140a0z

with normal mapping front:
http://cl.ly/image/2E3X0Y3b1s2c

with normal mapping back:
http://cl.ly/image/2g2Y1D3n360P

As you can see here also on the specular only picture (from the back of the cube):
http://cl.ly/image/0d0G3P0s252T

The NdotL product here is wrong as it can not possibly be > 0.0f.
Everything except the front face of the cube should essentially be lit and the rest completely black as it is the case when not using normal mapping.
All sides of the cube seem to have a final normal that does make the side faces lit a little bit except the face in the back, which makes the whole image look odd.
Why is this ? Is there some angle check that I'm missing ?

Here's how I calculate the tangent/bitangent in my application:
// Calculate bitangent
   for(int i = 0; i < 24; i++)
   {
	XMVECTOR normal = XMVectorSet(this->texturedNormalVertices[i].normal.x,
			 this->texturedNormalVertices[i].normal.y,
			 this->texturedNormalVertices[i].normal.z,
			 0);
	XMVECTOR tangent = XMVector3Cross(normal, forward);
	XMFLOAT4 tang;
	XMStoreFloat4(&tang, tangent);
	float magnitude = sqrt(tang.x * tang.x + tang.y * tang.y + tang.z * tang.z);
	if( magnitude == 0 )
	{
	 tangent = XMVector3Cross(normal, up);
	}
	XMVECTOR biTangent = XMVector3Cross(normal, tangent);
	// Calculate tangent
	XMVECTOR dot = XMVector3Dot(XMVector3Cross(normal, tangent), biTangent);
	if (XMVectorGetX(dot) < 0.0f)
	{
	 tangent = tangent * -1.0f;
	}
	XMStoreFloat3(&this->texturedNormalVertices[i].tangent, tangent);
	XMStoreFloat3(&this->texturedNormalVertices[i].biTangent, biTangent);
   }

And here's the shader code:
VS part:
// Calculate View-space Tangent-BiTangent-Normal matrix
output.TBN[0] = mul(input.Tangent, (float3x3)WorldView);
output.TBN[1] = mul(input.BiTangent, (float3x3)WorldView);
output.TBN[2] = mul(input.Normal, (float3x3)WorldView);

PS part:
// Normal Mapping
float4 NormalMapped = (2.0f * NormalMap.Sample(LinearSampler, input.UV)) - 1.0f;
NormalMapped = normalize(NormalMapped);
float3 N = normalize(mul(NormalMapped.xyz, (float3x3)input.TBN));

the BRDF:
float4 Gaussian(in float3 Normal, in float3 LightDir,
	in float3 ViewDir, in float SpecularPower,
	in float3 SpecularAlbedo, in float SpecularIntensity)
{
const float kPI = 3.1415926536f;

float3 N = normalize(Normal);
float3 L = normalize(LightDir);
float3 V = normalize(ViewDir);
float3 H = normalize(-L - V);

float4 Lr = float4(0, 0, 0, 1);
float4 Ls = float4(0, 0, 0, 1);

// Incident irradiance
float4 Ei = _LightColor * _LightIntensity;

// Lambert
float NdotL = saturate(dot(N, -L));
float kd = NdotL; // diffuse constant
Lr = (kd / PI) * Ei; // outgoing radiance

// Gaussian Specular [Mittring12]
float normalizationFactor = 0.17287429f + 0.01388682f * SpecularPower; // approximation by Krzysztof Narkowicz
float NdotH = saturate(dot(N, H));
float Threshold = 0.04f;
float cosAngle = pow(Threshold, 1 / SpecularPower);
float normAngle = (NdotH - 1) / (cosAngle - 1);

float Specular = exp(-normAngle * normAngle) * NdotL;
Specular *= normalizationFactor;

Ls = float4(Specular, Specular, Specular, 1.0f);
Ls.rgb *= SpecularAlbedo * SpecularIntensity;

return float4(Lr + Ls);
}

Edited by lipsryme, 11 October 2012 - 09:18 AM.


Sponsor:

#2 kauna   Crossbones+   -  Reputation: 1889

Like
0Likes
Like

Posted 12 October 2012 - 07:48 AM

Hi,

Have you tried a simpler NdotL lighting model? Have you verified that all your calculations done in the view space?
Have checked that your TBN matrix is calculated correctly?

Cheers!

#3 lipsryme   Members   -  Reputation: 878

Like
0Likes
Like

Posted 12 October 2012 - 11:17 AM

Ah... I knew the brdf would distract people Posted Image
It shouldn't be wrong. Plus I've tested it with just a saturate(dot(N, -L)) as output with the same result.
If I don't use the TBN matrix and just use the normals everything works as intended. But once I combine that with the input from the normal map and the TBN matrix it goes wrong.

Well how much more is there to it than just this?

// Calculate View-space Tangent-BiTangent-Normal matrix

output.TBN[0] = mul(input.Tangent, (float3x3)WorldView);
output.TBN[1] = mul(input.BiTangent, (float3x3)WorldView);
output.TBN[2] = mul(input.Normal, (float3x3)WorldView);


I'm just taking the rotational part of the worldview matrix (yes I've tried just world space lighting -> same issue) and multiplying that with the tangent, bitangent and normal for this specific vertex. Basically just what I do in my code snippets up there.

update: so pix is telling me I've got an NdotL of 0.436 when looking at the side of the cube.
LightDirection in view space = (-0.997, -0.002, 0.074)
Normal in view space = (-0.368, -0.055, 0.928)

The Light Direction is most definitely right so it's obviously that the problem seems to be the normal.
The thing though is like I said it's just wrong when using the TBN. So either the tangent/bitangent is wrong or something I'm doing in the shader code.

update2: looking at the vertex shader I can see the following:
input.Tangent = (0, 1, 0)
input.BiTangent = (0, 0, -1)
input.Normal = (-1, 0, 0)

TBN after mul with WorldView matrix:
(0, 0.250, 0.007)
(0.249, 0, -0.018)
(-0.018, 0.007, -0.249)


update3: I just tried normal mapping with the same textures in FXcomposer. Funny thing is the result is the same.
But I don't really get it. I don't see how this is realistic. Having almost half the light that shines onto the front face also on the side faces and the back face of the cube being completely black.

Edited by lipsryme, 13 October 2012 - 10:46 AM.





Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS