View more

View more

View more

Image of the Day Submit

IOTD | Top Screenshots

The latest, straight to your Inbox.

Subscribe to GameDev.net Direct to receive the latest updates and exclusive content.

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.

6 replies to this topic

#1lipsryme  Members

Posted 30 April 2013 - 10:19 AM

Hey guys, I've been trying to implement a phsically based metal kind of shader and I was wondering if what I'm doing is correct.

Have a look at my shader:

	// Compute half-way-vector
float3 H = (V + L);
H = normalize(H); // normalize

// Specular Fresnel Term approximation by Schlick
float LdotH = saturate(dot(L, H));
float3 Fresnel_spec_multiplier = 1.0f - FresnelReflectance;
float3 Fresnel_spec = FresnelReflectance + exp2(log2(1.0f - LdotH) * 5 + log2(Fresnel_spec_multiplier));

// Normalized Blinn-Phong NDF (D)
float D = 0.0f;
float NdotH = saturate(dot(N, H));
float NormalizationFactor = ((glossiness + 2) / 2);
D = exp2(log2(NdotH) * glossiness + log2(NormalizationFactor)); // NormalizationFactor * pow(NdotH, glossiness)

// Cook-Torrance geometry term approximation by Kelemen & Szirmay-Kalos
float G = 1.0f / (LdotH * LdotH + 0.0000001f);

// Microfacet BRDF f(l,v)
float3 f_microfacet = (Fresnel_spec * G * D) / 4.0f;

Radiance += f_microfacet * Li * Cosine;


As you can see I'm already using a custom microfacet brdf using an approximation of the cook-torrance geometry term and the fresnel factor but for the NDF just a regular blinn-phong model. Is the cook-torrance NDF actually the beckmann distribution function or is it something different ? If yes then should I rather use a precomputed distribution texture for this like it's found on here: http://http.developer.nvidia.com/GPUGems3/gpugems3_ch14.html ?

I've found this code segment for the cook-torrance NDF on pixars renderman page:

float alpha = acos(NdotH);
float D = gaussConstant*exp(-(alpha*alpha)/(m*m));


alpha in this case would be NdotH and m is the microfacet H ? And the guassian constant would be something like 0.8346...? (according to what I found on wikipedia).

This is a screenshot of how it looks using fresnel of gold:

http://d.pr/i/mOBL

here's another one using a different normal map:

http://d.pr/i/xwz7

I think it's wrong to tint the environment color with the specular fresnel again after I've multiplied it before by the fresnel using NdotV ?

It's just that it looks very "blueish" to me if I don't do that, have a look:

http://d.pr/i/RiCt

Edited by lipsryme, 30 April 2013 - 10:26 AM.

#2LouisCastricato  Members

Posted 30 April 2013 - 12:25 PM

You could just tint the environment color. There isn't much of an issue with that.

#3Chris_F  Members

Posted 30 April 2013 - 12:43 PM

You shouldn't be multiplying your environment map with Fresnel_spec, which uses the half vector.

I would not recommend Gaussian distribution function, or Beckmann, or texture look-ups. GGX is about as cheap as Blinn-Phong to compute, and supports very rough surfaces like Beckmann. Out of all of them I think it looks the best too.

As for your gold looking "blueish", of course it does. It's in a blue environment. If everything around it is blue, then what else is there to reflect but blue?

#4lipsryme  Members

Posted 30 April 2013 - 12:54 PM

Well yeah...was just not sure if it would be this strong of a blue tone since gold would be kind of yellowish I guess (the specular itself is more so, so I guess its okay...)

I removed the multiplication, I realized it was wrong since I was multiplying it two times in a row with the fresnel term.

About the GGX, do you have a sample code / paper on how to implement that ?

#5Chris_F  Members

Posted 30 April 2013 - 01:16 PM

Well yeah...was just not sure if it would be this strong of a blue tone since gold would be kind of yellowish I guess (the specular itself is more so, so I guess its okay...)

I removed the multiplication, I realized it was wrong since I was multiplying it two times in a row with the fresnel term.

About the GGX, do you have a sample code / paper on how to implement that ?

float NdotH_2 = NdotH * NdotH;

float a_2 = a * a;

float D = a_2 / pow(NdotH_2 * (a_2 - 1.0) + 1.0, 2.0);


#6lipsryme  Members

Posted 30 April 2013 - 01:25 PM

Thanks a lot. Is the roughness input value range similar to blinn-phong ? Meaning 0-2048 (low to high gloss)

#7Chris_F  Members

Posted 30 April 2013 - 01:37 PM

Thanks a lot. Is the roughness input value range similar to blinn-phong ? Meaning 0-2048 (low to high gloss)

No, it's similar to Beckmann (0, +inf], but you will probably stick to values in the range of 0-1 for most common materials.

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.