Jump to content

  • Log In with Google      Sign In   
  • Create Account


How can i have a softer Skin using BRDF


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
19 replies to this topic

#1 toto5100   Members   -  Reputation: 169

Like
1Likes
Like

Posted 24 June 2014 - 02:28 PM

Hi,

i implemented physically based shading, using the Cook Torrance BRDF Equation, with Trowbridge Reitz GGX for Normal Distribution, the Smith Schlick GGX for Geometric Shadowing, Schlick Approximation for Fresnel and for diffuse, Lambert.

All is working in Linear Space with pow(2.33) on Input and pow(0.4545) on output, and it looks good.

But the problem i have is for skin :

Skin harsh.jpg

The falloff is too harsh, good for general objects, but the skin should look softer.

And it looks like cell shading...

So how can i make it softer like this :

crysis3-2013-01-22-10-57-22-06.jpg



Sponsor:

#2 Promit   Moderators   -  Reputation: 6614

Like
9Likes
Like

Posted 24 June 2014 - 03:24 PM

You probably want to look into subsurface scattering for the skin. Jimenez's screen space shader is the popular one. However, I'm concerned about how sharp your light-shadow transition is independent of the material being rendered. That does not look correct to me. You should check out what's going on there before doing anything else. FWIW, a cheap starting point with skin is to simply shut off specular entirely. This is a good sanity check.


Edited by Promit, 24 June 2014 - 03:25 PM.


#3 kauna   Crossbones+   -  Reputation: 2338

Like
3Likes
Like

Posted 24 June 2014 - 04:29 PM

Why aren't you taking advantage of gamma corrected texture reads / back buffers?

 

Cheers!



#4 MJP   Moderators   -  Reputation: 10909

Like
6Likes
Like

Posted 24 June 2014 - 06:50 PM

You probably want to look into subsurface scattering for the skin. Jimenez's screen space shader is the popular one. However, I'm concerned about how sharp your light-shadow transition is independent of the material being rendered. That does not look correct to me. You should check out what's going on there before doing anything else. FWIW, a cheap starting point with skin is to simply shut off specular entirely. This is a good sanity check.

 

I think it looks right for a gamma-correct Lambertian diffuse falloff. Linear-space Lambertian is actually pretty harsh, especially compared to doing it in gamma-space. Obviously it looks totally wrong for skin, since the SSS will "soften" the falloff and avoid that "rocky" look that you would get otherwise.

So yeah, there's the screen-space stuff which has its pros and cons. At work we use Pre-Integrated Skin Shading (also has an article in GPU Pro 2), which has a different set of pros and cons. For some background reading, I would also suggest reading the skin rendering chapter from GPU Gems 3. The actual techniques it describes aren't really attractive for real-time rendering (texture-space diffusion is too expensive and doesn't scale well, and GGX is a cheaper approximation than multiple Beckmann specular lobes), but it's still a great article.


Edited by MJP, 06 July 2014 - 03:42 PM.


#5 Tutorial Doctor   Members   -  Reputation: 1564

Like
0Likes
Like

Posted 24 June 2014 - 09:39 PM

I think I accidentally down voted someone in this post. Tablet jumping around.

They call me the Tutorial Doctor.


#6 toto5100   Members   -  Reputation: 169

Like
0Likes
Like

Posted 25 June 2014 - 02:12 AM

@Promit Yes, the falloff is very harsh, and without specular too (it stay the same). Do you know another diffuse equation i can use to have something softer ?

@kauna It's because Unity Free have his inputs in Gamma Space, so i had to manually correct the gamma in shaders.

 

I will try Sub Surface Scaterring, but i cant' use a method in Screen Space or anything which needs a rendertexture : in free version, rendertexture are very expensive.

So i'll try the Pre-Integrated Skin Shading but i have some questions :

if i understood well, i need my usual textures :

- Albedo, Normal map, Specular Map, Gloss Map

and

- Curvature Map (Baked from the mesh) and BRDF look up

So this method only affect diffuse calculations right ?

I can keep my Cook Torrance Specular, i just have to modify classic Diffuse Lambert to add SSS ?



#7 Tutorial Doctor   Members   -  Reputation: 1564

Like
0Likes
Like

Posted 25 June 2014 - 10:59 AM

Will sub-surface scattering necessarily help the falloff though? You can however get good skin results with nice ambient occlusion, SSS, a subtle texture map, and a decent normal map.

They call me the Tutorial Doctor.


#8 Promit   Moderators   -  Reputation: 6614

Like
1Likes
Like

Posted 25 June 2014 - 11:06 AM


@Promit Yes, the falloff is very harsh, and without specular too (it stay the same). Do you know another diffuse equation i can use to have something softer ?
At risk of revealing my mountain man situation of being trapped on mobile ... I just like gamma space half lambert, as described in the HL2 slides.
	float ndotl = dot(N, L1);
	//compute half lambert term
	ndotl = (ndotl * 0.5) + 0.5;
	ndotl *= ndotl;
	ndotl = clamp(ndotl, 0.0, 1.0);



#9 toto5100   Members   -  Reputation: 169

Like
0Likes
Like

Posted 25 June 2014 - 12:17 PM

I tested your half lambert, but the problem is the lighting looks flat : the falloff is very large, so it looks incorrect, especially with shadows.

I tried to downscale the normal map, and the result is a little softer, but it's still very harsh.

Image based Lighting Woman Harsh 2.jpg

I'll try pre integrated sss too.

Someone knows where i can download the head scan all people use ? The Infinite site is down :(



#10 MJP   Moderators   -  Reputation: 10909

Like
4Likes
Like

Posted 25 June 2014 - 03:14 PM

So this method only affect diffuse calculations right ?

 

 

Yes. Current sub-surface scattering techniques only address diffuse reflectance, so you can leave specular alone.

 

 

 

Will sub-surface scattering necessarily help the falloff though?


Absolutely. Take a look at this image from Eric Penner's slides:

 

SSS.PNG

 

Notice how the falloff becomes much softer as curvature increases.



#11 Promit   Moderators   -  Reputation: 6614

Like
0Likes
Like

Posted 25 June 2014 - 10:20 PM

Are you using self shadowing in these screenshots?



#12 toto5100   Members   -  Reputation: 169

Like
0Likes
Like

Posted 04 July 2014 - 03:24 PM

Hi,

Now, i implemented Pre Integrated SSS, so i use curvature and BRDF lookup but i have a new problem :

the SSS have ugly artifacts, and i don't know why. I tried to use a smaller mip map but it doesn't resolve the problem.

1404506798-bad-sss.jpg

You can see the SSS is pretty hard, with a "polygon" look.

 

After that, i tried to do a pow(2.2) on the lookupsample, and there is less artifacts but it's incorrect(the falloff is not linear).

I also have some banding.



#13 L. Spiro   Crossbones+   -  Reputation: 13268

Like
0Likes
Like

Posted 04 July 2014 - 07:07 PM

The look-up table is data, not imagery, and should not be “converted” to any spaces (including linear).

 

It would appear that you have a problem with normals.  How does the model render normally?

 

 

L. Spiro


It is amazing how often people try to be unique, and yet they are always trying to make others be like them. - L. Spiro 2011
I spent most of my life learning the courage it takes to go out and get what I want. Now that I have it, I am not sure exactly what it is that I want. - L. Spiro 2013
I went to my local Subway once to find some guy yelling at the staff. When someone finally came to take my order and asked, “May I help you?”, I replied, “Yeah, I’ll have one asshole to go.”
L. Spiro Engine: http://lspiroengine.com
L. Spiro Engine Forums: http://lspiroengine.com/forums

#14 Chris_F   Members   -  Reputation: 2227

Like
0Likes
Like

Posted 04 July 2014 - 07:37 PM

Looks to me like your curvature map is simply rubbish.



#15 toto5100   Members   -  Reputation: 169

Like
0Likes
Like

Posted 05 July 2014 - 10:03 AM

It's not a problem of normal :

1404570573-normal-good.jpg

Normals are ok.

 

And about the curvature, i calculate it dynamically using :

length( fwidth(WorldNormal)  ) / length( fwidth (worldPos ) );

 

I'm manually correcting gamma by doing pow(2.2) on input (Albedo) and pow(1/2.2) on output.

When i remove this (so without gamma correction) it looks like it should (but its in gamma).

 

So i found a compromise which is i do gamma correction AFTER brdf calculations.

So the code looks like this :

half3 falbedo = tex2D(_MainTex,uv);
float ndotl = dot(s.blurNormal,lightDir);
lookUpPos.x = ndotl * 0.5 + 0.5;
lookUpPos.y = curvature * lightIntensity;
float3 dif = tex2D (_BRDFLookUp, lookUpPos).xyz;
fAlbedo*=dif;
float3 final = (pow(fAlbedo,2.2) + spec) * _LightColor0.rgb * (attenuation * 2);
return float4(final,1);

I know that it's not correct but i found nothing else.


Edited by toto5100, 05 July 2014 - 10:04 AM.


#16 Tasty Texel   Members   -  Reputation: 1253

Like
0Likes
Like

Posted 06 July 2014 - 04:13 AM

If you calculate your curvature map using screen space derivatives of lerped normals then it is only to be expected that the curvature values are constant over individual polygons. Normalizing your normals beforehand could help a bit, I'm not sure how much though.

 

EDIT: Ok, the prob is that you have to consider the change in position too, which is constant over triangles anyways.


Edited by Tasty Texel, 06 July 2014 - 04:26 AM.


#17 toto5100   Members   -  Reputation: 169

Like
0Likes
Like

Posted 06 July 2014 - 06:29 AM

When i do :

length( fwidth(normalize(WorldNormal) )  ) / length( fwidth (worldPos ) );

it changes nothing.

Also, i'm using Unity's Surface Shaders, which compute normals automatically. Normal map is also taken in account.

I don't know how can i smooth them. It's also strange that the SSS effect works well when i do BRDF before gamma correction and looks ugly when i do gamma correction on input.

 

PS : You think Cook Torrance with GGX is good for skin specular or i should move to Kelemen/Szirmay-Kalos ?


Edited by toto5100, 06 July 2014 - 06:45 AM.


#18 MJP   Moderators   -  Reputation: 10909

Like
1Likes
Like

Posted 06 July 2014 - 03:41 PM

In my experience you really need to use curvature maps for the pre-integrated shading. The trick with using derivatives and clever and might work for certain simple scenarios, but the fact it has discontinuities across triangle edges is pretty much a deal-breaker for faces.

 



#19 Tasty Texel   Members   -  Reputation: 1253

Like
0Likes
Like

Posted 06 July 2014 - 04:28 PM

@MJP: Do you use a single curvature map regardless of facial animations?



#20 MJP   Moderators   -  Reputation: 10909

Like
2Likes
Like

Posted 07 July 2014 - 08:16 PM

Yeah we always use the same map, we don't animate it.






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