Implement indirect specular occlusion in StandardMaterial3D

[Calinou]

master version of #50603.

Both sources of AO (SSAO and AO texture maps) are used for this effect. The difference is most noticeable when SSAO is enabled (especially with height maps), but this effect also helps when SSAO is disabled.

It is an approximate effect with a very low performance cost, but makes specular reflections more believable overall.

Testing project: test_ao_specular_occlusion_master.zip

Preview

SSAO intensity was doubled to make the difference more noticeable. In real world scenarios, the difference will be more subtle.
Light Affect is set to 0 (default) both in the SSAO and material settings.

Left: partially metallic material (0.5), right: fully metallic material (1.0)

SSAO and height enabled

Forward+

Before	After

Mobile¹

Before	After

No SSAO and no height (only AO map in material)

Forward+

Before	After

Mobile¹

Before	After

Footnotes

1. This currently uses a faster, but supposedly more approximate approach to indirect specular occlusion. ↩ ↩²

[clayjohn]

Looks great as a start!

However, I don't think we should force light_affect to 1.0 for specular lighting. Light_affect is already kind of a specific hack that will only be used by some, I don't think we should force it on all users with no way to turn off.

An alternative is to separate direct specular from indirect specular and then combine them after you have applied AO to specular_light. Separating the two will also probably be desirable as we polish other rendering features.

[Calinou]

Rebased and tested again, it works as expected. New screenshots were added to OP, along with a testing project.

Let me know if I should add an Environment property to adjust specular AO intensity, though I'd prefer the effect to be enabled by default. If the unused Ao Channel Affect Environment property is removed, this should free up some space in the Environment inspector anyway.

[clayjohn]

This PR still has the problem that it treats direct and indirect specular the same. It is darkening both direct and indirect by the full amount of ao regardless of the ao_light_effect setting. AO should only impact indirect light unless a user specifically sets it to impact direct lighting. The Filament docs have a very nice section explaining how specular occlusion can be derived from ambient occlusion: https://google.github.io/filament/Filament.html#lighting/occlusion/specularocclusion

[Calinou]

This PR still has the problem that it treats direct and indirect specular the same. It is darkening both direct and indirect by the full amount of ao regardless of the ao_light_effect setting. AO should only impact indirect light unless a user specifically sets it to impact direct lighting. The Filament docs have a very nice section explaining how specular occlusion can be derived from ambient occlusion: google.github.io/filament/Filament.html#lighting/occlusion/specularocclusion

I managed to separate direct specular lighting from indirect specular lighting and use the Filament approximation. See OP for updated screenshots 🙂

However, this appears to have a significant performance cost (0.2 ms of added frame time in my test scene, with SSAO and height mapping disabled). I expected it to be cheaper, given the description on the Filament website. I've tried to move ndotv so it's only calculated once, but performance didn't get better.

If I remove the Filament approximation but keep the indirect specular multiplication based on original AO, I get the exact same result (but with better performance):

SSAO + height

No SSAO and height

Difference with original "After" images

The images are bit-for-bit identical:

dssim 2022-02-07_23.20.15_highend_after_ao.png 2022-02-07_23.37.55_highend_after_ao_cheaper.png 
0.00000000      2022-02-07_23.37.55_highend_after_ao_cheaper.png

dssim 2022-02-07_23.20.38_lowend_after_ao.png 2022-02-07_23.35.49_lowend_after_ao_cheaper.png 
0.00000000      2022-02-07_23.35.49_lowend_after_ao_cheaper.png

Note that I still need to port the changes to the mobile renderer.

[Calinou]

Rebased and tested again, it works as expected. I've updated screenshots in OP. Note that this comment still applies.

[Calinou]

• Superseded by Add specular occlusion term from AO and Bent Normals and use Bent Normals for indirect lighting #89988.