Sheen E LUT is not properly initialized

[taylor-santos]

I was examining the sample viewer's output in RenderDoc and noticed that the Sheen E LUT texture is just a single 1x1 black pixel:

The expected texture is included in the glTF-Sample-Renderer repo under assets/images/lut_sheen_E.png and looks like this:

The texture is sampled by the fragment shader in the albedoSheenScalingLUT function:
source/Renderer/shaders/material_info.glsl:431

float albedoSheenScalingLUT(float NdotV, float sheenRoughnessFactor)
{
    return texture(u_SheenELUT, vec2(NdotV, sheenRoughnessFactor)).r;
}

And examining the disassembled version of this line in RenderDoc's debugger we can see the sampled value (r0.z) is 0.00:

The texture is supposed to be initialized here:
source/ResourceLoader/resource_loader.js:379:

environment.images.push(new gltfImage(luts.lut_sheen_E_file, GL.TEXTURE_2D, 0, undefined, undefined, ImageMimeType.PNG));
const sheenELut = new gltfTexture(lutSamplerIdx, [imageIdx++], GL.TEXTURE_2D);
sheenELut.initialized = true; // iblsampler has already initialized the texture
environment.textures.push(sheenELut);

Possible Solution:

Setting sheenELut.initialized = false instead, I can see that the texture is now loaded properly in RenderDoc:

And we now see a non-zero sample in the debugger:

With this fix applied, we see a very slight change in the appearance of materials using KHR_materials_sheen. Here are some side-by-side comparisons, ranging from most obvious to least:

[andrewvarga]

I can verify this with babylon.js's Spector (webgl inspector), charlieLUT texture seems initialized but sheenELUT seems not:

Somewhat related, but this line is also confusing a bit:
https://github.com/KhronosGroup/glTF-Sample-Renderer/blob/95ad8b4d2067e53337d34089c0ee06159c2ab4d3/source/ResourceLoader/resource_loader.js#L384

sheenLUT and ggx lut texture infos explicitly set the linear parameter to true a few lines above:

environment.lut = new gltfTextureInfo(environment.textures.length - 1, 0 , true);

but sheenELUT doesn't. The default is true so it's the same result but I think they should all either explicitly set it or none of them.