Charles Thomas's Website

Skin shading

21/04/2016

In the past I managed to use Jorge Jimenez separable sub surface scattering^[1] on Unity 4.

At the time I couldn't use the deferred (pre-pass) of Unity 4 so I was stuck in forward.

I had to use a second camera in order to get the diffuse and specular lighting pass. It was of course a bit slow and even slower if too much character was in the scene but it was working.

I got back to this recently for a contract work but this time for Unity 5.

Top condition was to use deferred but in Unity 5 you can't have access to a diffuse and specular lighting pass so I had to find a way to get them, a second camera was not an option.

While working for a personnal project, trying to change how data are packed into the g-buffer in Unity 5 using interleaved YCbCr^[2]^[3] this leads me to how it's done in Unreal Engine 4^[4].

Basically you pack the diffuse and specular lighting with a squared interleaved pattern. Then you can seperate them using that same pattern, but you'll have to find a way to retreive the missing data (empty square).

The process works as follow :

1 - Diffuse and specular g-buffer remain the same, I'll need them later at the final lighting reconstruction pass.

2 - Deferred lighting pass is modified in order to pack the diffuse and specular lighting using the squared interleaved pattern.

Diffuse/Specular lighting.

3 - In a post process shader diffuse and specular lighting pass are separated and stored at half the screen resolution in two render texture.

Diffuse lighting.

Specular lighting.

4 - A 2x2 bicubic filter is applied on both diffuse and specular lighting pass.

5 - SSSSS pass on the diffuse lighting (Notice the edge bleeding).

SSSSS pass.

6 - Final lighting reconstruction with the diffuse color multiply with SSSSS diffuse lighting and specular added on top (DiffuseColor * SSSSSDiffuseLighting + SpecularLighting).

Close up of the final shading with varying scattering.

Results are pretty good but shows edge bleeding and at far distance the pixels are very jaggy (due to the half resolution) but overall it's a huge win compared to what I've done in the past.

I'm also using interleaved gradient noise^[5]. The resulting image is much smoother and if you have a Temporal AA you can animate it for an ever better look.

Thanks for reading and if you have questions I'll be happy to answer them.

Head scan is from: http://ten24.info.

References

[1] https://github.com/iryoku/separable-sss
[2] The Compact YCoCg Frame Buffer
[3] The Rendering Technologies of Crysis 3
[4] Digital Humans
[5] Next Generation Post Processing In Call Of Duty Advanced Warfare