[WIP] PBR Materials extension

[mlimper]

This is our (currently WIP) PBR material extension - this PR is currently mainly intended to serve as a place for discussion.

Direct Preview:
https://github.com/tsturm/glTF/tree/master/extensions/Vendor/FRAUNHOFER_materials_pbr

[pjcozzi]

@erich666 would you or any of your colleagues be interested in reviewing this proposed PBR materials extension to glTF?

[pjcozzi]

You'll want this to be using the 1.0.1 spec, which may require no changes to the extension. We plan to finish 1.0.1 for SIGGRAPH. See #605.

[pjcozzi]

I don't have experience with PBR materials, but I just want to confirm that it is common to use the alpha channel of the diffuse factor for opacity.

[pjcozzi]

One of my students, @jian-ru, added support for the current version of this extension to his Vulkan engine: https://github.com/jian-ru/laugh_engine#gltf-support

[pjcozzi]

Also WIP WebGL glTF PBR reference implementation (raw WebGL, no engine) by @moneimne is here: http://www.seas.upenn.edu/~moneimne/WebGL-PBR/

Repo: https://github.com/moneimne/WebGL-PBR

[AurL]

From #817

We have also decided that the metallic-roughness PBR material model will be in the core glTF spec (specular-glossiness will remain in the extension)

Is there any material schema example that shows the diff between a metal-roughness material vs a specular-glossiness one ?

It seems to be a small change regarding the schema but as I am working on an exporter using PBR, I would like to be sure that it's done well. From what I understand, only the specular-glossiness material description would be under an EXT_materials_PBR extension ?

[pjcozzi]

@AurL the JSON schemas are here:

• materials_pbr_metal_roughness.schema.json
• materials_pbr_specular_glossiness.schema.json

@mlimper other than moving metallic-roughness into core spec, are these fully up-to-date?

[mlimper]

other than moving metallic-roughness into core spec, are these fully up-to-date?

Three points coming to my mind right now:

• @sbtron, @bghgary and colleagues suggested to make some additional comments, such as specifying priorities for the different maps, which should help a client that does not support the required number of channels.

• There should be the possibility to use an F0 texture, to specify F0 values for non-metallic parts. That case is a bit special, but we should cover it.

• I'll provide a separate document for additional maps (normals, occlusion), and I would like to link to that one in both specs, materials_common and materials_pbr.

The last point will take some more days, but the first two points can be done relatively quickly.

@bghgary Would it be possible for you and your colleague (Sebastien, I believe) to specify the prioritization of the different maps as we discussed it and open a PR for that?

[pjcozzi]

If there is anything @moneimne can do to help finish the spec/extension, please let us know.

[bghgary]

Yes, though for the PR, I would expect this information to go with the separate document for additional maps that you mentioned. I can wait until you are done or you can just incorporate this into your document directly.

List of additional maps in priority order:

Map	What happens if it’s not there
Tangent Space Normal	Geometry will appear less detailed than authored
Ambient Occlusion	Model will appear brighter in areas that should be darker
Emission	Model with lights will not be lit (e.g. the headlights of a car model will be off instead of on)
F0 (for dielectrics)	Some non-metallic materials, such as ruby and diamond, will be less shiny than it’s supposed to be

Resource-bound implementations should drop maps from the bottom to the top.

[mlimper]

Yes, though for the PR, I would expect this information to go with the separate document for additional maps that you mentioned. I can wait until you are done or you can just incorporate this into your document directly.

Yes, I'll provide that one. Please feel free to open the PR and already incorporate the table into the spec right now.

[javagl]

@selim-bekkar-sb mentioned in #643 (comment)

Why do you force to merge the Specular and Glossiness and or the Metallic and Roughness textures ?

The answer by @pjcozzi in #643 (comment) was:

The spirit of glTF is that it is easy and fast for an engine to load and render; it does not need to match how a modeling tool wants to store data since it is a "final stage" format that is intended for engines, not modeling tools.

Is it safe to assume that combining these textures will be enforced?

I'm asking because, for example, the "Helmet" test model and the work in https://github.com/moneimne/WebGL-PBR in general seem to operate on separate textures. I'd like to create a "baseline" version (not an "official reference" - only for personal use - but to be published at https://github.com/javagl/gltfTestModels/tree/master/DamagedHelmetModified ), and now assume that I'd create a combined version of the respective textures.

[mlimper]

Is it safe to assume that combining these textures will be enforced?

Yes, and also agreed with the comment of @pjcozzi

Imagine we would allow both, we would have the risk that each combination "exporter/renderer" implements only one of the possible ways, which would lead to assets not being fully interchangeable across engines

@moneimne
you probably used the separate textures as an intermediate step towards the final version of the demo, I guess?

[javagl]

So then I hope that https://github.com/javagl/gltfTestModels/blob/master/DamagedHelmetModified/glTF/Default_MetalSmooth_metallicRoughness.jpg can serve as a basis for further tests.

[moneimne]

you probably used the separate textures as an intermediate step towards the final version of the demo, I guess?

Yes, I have been using the Helmet test model as is for now, but I can definitely use the combined version of the model once the changes have been made.

Thank you for the combined texture, @javagl! Do you also plan on updating the glTF file to reflect this change?

[javagl]

Sure. I just started working on all that, and intended to create a shader implementation that implements the spec as closely as possible. I started reading the Frostbite PBR document, but it's difficult to flesh out the relevant parts here. E.g. the spec does not cover normals/bump maps or ambient occulsion, and I still have to figure out how they are integrated eventually (there are some open issues for the discussion about these points).

In the meantime, I have updated the glTF and did some minor updates to the shaders, to use the combined metallic+roughness texture, and to take the factors (base color, metallic, roughness) into account: javagl/gltfTestModels@6682793

[pjcozzi]

Metal Roughness is being added to the core glTF 2.0 spec in #830.

[tparisi]

@pjcozzi can we have a succinct recap of the logic behind putting specular-glossiness into an extensions? @cedricpinson how does the decision relate to Sketchfab's PBR model out of curiosity?

[pjcozzi]

@tparisi this was discussed at length with broad community input, see #696 (comment)

[cedricpinson]

@tparisi well nothing special we adapted our implementation to fit the extension. I would have been happy to have this in the core, but it does not really change a lot. I understand that simplicity with metalness,roughness is good for everybody.

[lexaknyazev]

Should we close this PR, or it will be published as glTF 1.0 extension?

[mlimper]

I think we could close it.

@pjcozzi What is your opinion?

[pjcozzi]

OK with me. If there are any TODOs in comments, please capture them in a separate issue.

[mlimper]

I believe everything that was at some point identified as a separate TODO has already been captured in separate issues, which are all mentioned in this thread.
The ones that are still open are:
#697
#699