Revert the change of the limit for interpolation of F0 for dielectrics and metals for Screen Space Reflections

[mandryskowski]

Commit 2c000cb changed the interpolation limits from (0.04, 1.0) to (0.04, 0.37). This is incorrect, as we want to have an F0 of 0.04 for dielectrics (materials with metalness of 0.0) and an F0 of 1.0 for metals. The Schlick approximation uses an F0 of 0.04 for all dielectrics.
The Fresnel effect does not occur with metals and their F0 has to be 1.0 (they fully reflect normally incident light). Note that the model only assumes metalness of 0.0 or 1.0 but in practice we will have intermediate values when sampling textures which is why interpolating is valid in this case (even if not physically correct).
Relevant to #79549

[clayjohn]

Tested locally:

4.0-stable

4.1-stable

This PR

[clayjohn]

Looks good to me. The results are much more convincing.

[n0tank3sh]

The Shlick term uses 0.04 because the air as Air has a refractive index of 1.5 which from equation ((1 - 1.5)/(1 + 1.5))^2 we get 0.04 and we 0.37 from this same equation using the refractive index of germanium about 4, I guess. That's how I got 0.37. Total reflection is the theoretical limit that cannot be achieved practically. By changing it to 1 we once again going to arrive at the same problem mentioned in #75302 (comment).

[clayjohn]

The Shlick term uses 0.04 because the air as Air has a refractive index of 1.5 which from equation ((1 - 1.5)/(1 + 1.5))^2 we get 0.04 and we 0.37 from this same equation using the refractive index of germanium about 4, I guess. That's how I got 0.37. Total reflection is the theoretical limit that cannot be achieved practically. By changing it to 1 we once again going to arrive at the same problem mentioned in #75302 (comment).

That issue was caused by the erroneous flipping of the normal's Y-coordinate. The F0 term only controls the blend between the computed m value and 1.0 which controls how much of the radiance map to include in the reflection.

Ultimately, we need to store F0 in the scene shader instead of metallic so we can do better than that approximation

[n0tank3sh]

I am not mentioning the main issue but the sub issue mentioned there in the comment. Which was there was sudden change in the metallic material. plus f0 which is 1 that is 100 percent reflection of incident ray is not possible practically.

[mandryskowski]

The Shlick term uses 0.04 because the air as Air has a refractive index of 1.5 which from equation ((1 - 1.5)/(1 + 1.5))^2 we get 0.04 and we 0.37 from this same equation using the refractive index of germanium about 4, I guess. That's how I got 0.37. Total reflection is the theoretical limit that cannot be achieved practically. By changing it to 1 we once again going to arrive at the same problem mentioned in #75302 (comment).

I think this is the right formula to use but I also believe that there are materials with F0 very close to 1.0 (e.g. silver has F0 = 0.97 with wavelength set to 0.7 µm which is red light and F0 = 0.93 with 0.4 µm which is violet).

Your calculations for germanium are correct but this material does not have the highest possible reflectance. It might seem so since it has such a high (the highest?) refractive index.

However, calculating F0 is a bit different with metals since they have a complex refractive index (n + ik). We apply the same formula but using the complex value, so ((1-n-ik)/(1+n+ik))^2.

The result of this calculation will be a complex value. We take the magnitude (modulus) of this value to get our F0.

[YuriSizov]

Thanks! And congrats on your first merged Godot pull-request!

[YuriSizov]

We decided against cherry-picking it for 4.1.x because it may cause issues in existing Godot 4.1 projects that have to deal with the current behavior. We prefer to preserve compatibility in such situations, even if it's a compatibility with workarounds. If someone needs this fix, upgrading to 4.2 is their option.