Cycles-Bake shows black when Displament input is used #40068

New Issue

Dalai Felinto · 2014-05-07T17:37:27+02:00

Dalai Felinto commented

2014-05-07 17:37:27 +02:00

Trying to bake this file shows black:
displacement_height.blend

This also affects bump nodes (see #40060)

I tried @brecht's suggestion below, but it shows white, as if there was no displacement at all (instead of showing the checkers pattern expected from the file).

I think the problem in that file is with the bump mapping, which fails due to missing ray differentials (sd.dP).
A quick way to solve that would be to add this after shader_setup_from_sample:

sd.dP.dx = sd.dPdu;
sd.dP.dy = sd.dPdv;

That's not quite right, really this should take the UV map used for baking and the image resolution into account, for that you could look at bake_set_vlr_dxyco in Blender Internal if you want, to see how you can get those differentials from a UV map. Getting that right would mainly help for OSL image texture lookups and advanced procedural textures that do antialiasing.

Now questions for @brecht:

Is this what the du_dx, du_dy, dv_dx, dv_dy were intended for?
If so shouldn't we calculate them in Blender, instead of Cycles?
Also, would that mean sd.dP.dx = make_float3(du_dx, du_dy, 1.0f) and sd.dP.dy = make_float3(dv_dx, dv_dy, 1.0f)?

Trying to bake this file shows black: [displacement_height.blend](https://archive.blender.org/developer/F87584/displacement_height.blend) This also affects bump nodes (see #40060) I tried @brecht's suggestion below, but it shows white, as if there was no displacement at all (instead of showing the checkers pattern expected from the file). >I think the problem in that file is with the bump mapping, which fails due to missing ray differentials (sd.dP). > A quick way to solve that would be to add this after shader_setup_from_sample: ``` sd.dP.dx = sd.dPdu; sd.dP.dy = sd.dPdv; ``` > That's not quite right, really this should take the UV map used for baking and the image resolution into account, for that you could look at bake_set_vlr_dxyco in Blender Internal if you want, to see how you can get those differentials from a UV map. Getting that right would mainly help for OSL image texture lookups and advanced procedural textures that do antialiasing. Now questions for @brecht: * Is this what the **du_dx**, **du_dy**, **dv_dx**, **dv_dy** were intended for? * If so shouldn't we calculate them in Blender, instead of Cycles? * Also, would that mean sd.dP.dx = make_float3(du_dx, du_dy, 1.0f) and sd.dP.dy = make_float3(dv_dx, dv_dy, 1.0f)?

Dalai Felinto commented

2014-05-07 17:37:27 +02:00

Changed status to: 'Open'

Brecht Van Lommel was assigned by Dalai Felinto

2014-05-07 17:37:27 +02:00

Dalai Felinto commented

2014-05-07 17:37:27 +02:00

Added subscriber: @dfelinto

Brecht Van Lommel commented

2014-05-07 18:50:42 +02:00

Yes, that's what the du_dx, du_dy, dv_dx, dv_dy were for. Calculating them in Blender would be best I think since otherwise Cycles has to have some knowledge about what you're baking to (uv mapped image, vertex colors, ptex, ..), and that's currently nicely separated, so best keep it that way.

The formula to compute dP.dx and dP.dy is:

dP.dx = dPdu * dudx + dPdv * dvdx;
dP.dy = dPdu * dudy + dPdv * dvdy;

To compute dudx you take the difference between u at the pixel to the right of the current pixel and the current pixel, and similarly for the others values.

Yes, that's what the du_dx, du_dy, dv_dx, dv_dy were for. Calculating them in Blender would be best I think since otherwise Cycles has to have some knowledge about what you're baking to (uv mapped image, vertex colors, ptex, ..), and that's currently nicely separated, so best keep it that way. The formula to compute dP.dx and dP.dy is: ``` dP.dx = dPdu * dudx + dPdv * dvdx; dP.dy = dPdu * dudy + dPdv * dvdy; ``` To compute dudx you take the difference between u at the pixel to the right of the current pixel and the current pixel, and similarly for the others values.

Dalai Felinto commented

2014-05-07 22:31:21 +02:00

Added subscriber: @brecht

Dalai Felinto commented

2014-05-07 22:31:21 +02:00

@brecht Cycles part is done - D510, I'm tackling Bake API part of it now. Feel free to review them separately.

@brecht Cycles part is done - [D510](https://archive.blender.org/developer/D510), I'm tackling Bake API part of it now. Feel free to review them separately.

Brecht Van Lommel commented

2014-05-08 16:34:12 +02:00

In this example file you're using the (experimental, not supported yet) displacement and subdivision feature. Displacement does not actually need ray differentials to work, only bump mapping needs it. The problem here is with the subdivision which alters the mesh topology and so the mapping to the original mesh is lost. If this used a subdivision surface modifier instead it would work.

The subdivision code at the moment doesn't even support things like preserving UV maps on meshes, it's unfinished, so I would not consider that a bug, there's a reason it's hidden behind that experimental option.

In this example file you're using the (experimental, not supported yet) displacement and subdivision feature. Displacement does not actually need ray differentials to work, only bump mapping needs it. The problem here is with the subdivision which alters the mesh topology and so the mapping to the original mesh is lost. If this used a subdivision surface modifier instead it would work. The subdivision code at the moment doesn't even support things like preserving UV maps on meshes, it's unfinished, so I would not consider that a bug, there's a reason it's hidden behind that experimental option.

Dalai Felinto commented

2014-05-08 19:28:57 +02:00

A simpler test file: displacement.blend

Look at the result with latest D510:

It seems scaled up 2x. Now if I connect the checkers texture in the diffuse node too:

A simpler test file: [displacement.blend](https://archive.blender.org/developer/F87896/displacement.blend) Look at the result with latest [D510](https://archive.blender.org/developer/D510): ![checker-displace.jpg](https://archive.blender.org/developer/F87897/checker-displace.jpg) It seems scaled up 2x. Now if I connect the checkers texture in the diffuse node too: ![checker-displace-diffuse.jpg](https://archive.blender.org/developer/F87899/checker-displace-diffuse.jpg)

blender-admin commented

2014-05-08 19:38:30 +02:00

This issue was referenced by blender/blender-addons-contrib@31cb7e395e

This issue was referenced by blender/blender-addons-contrib@31cb7e395e656e4a1590e1c9c4531e8a4b4f32ea

blender-admin commented

2014-05-08 19:38:30 +02:00

This issue was referenced by 31cb7e395e

This issue was referenced by 31cb7e395e656e4a1590e1c9c4531e8a4b4f32ea

blender-admin commented

2014-05-08 19:38:43 +02:00

This issue was referenced by blender/blender-addons-contrib@672cebf19a

This issue was referenced by blender/blender-addons-contrib@672cebf19a8b7fe8f9ac0f732fd6e8a0df239c7f