Zero sized normals if the object scale to zero in one axis. #74317

New Issue

Shish · 2020-02-29T15:32:30+01:00

Shish commented

2020-02-29 15:32:30 +01:00

System Information
Operating system: Darwin-19.3.0-x86_64-i386-64bit 64 Bits
Graphics card: AMD Radeon Pro 555X OpenGL Engine ATI Technologies Inc. 4.1 ATI-3.5.5

Blender Version
Broken: version: 2.82 (sub 7), branch: master, commit date: 2020-02-12 16:20, hash: 77d23b0bd7

A 3D object with a reflective texture turns black when flattened by scaling x0 along an axis

~

I was rendering a transform of a reflective hemisphere from concave to convex, seeing how the reflection changes over time.

With Cycles, this works as expected. Concave is warped one way, flat is a flat mirror, convex is warped the other way.
With Eevee, the concave and convex reflections are fine, but for one frame where the hemisphere has become a 2D circle, it shows up black.

Example file attached, saved at the half-way point in the animation where this bug occurs: spoon.blend

Videos attached to show the effect just in case this is hardware-specific and only happens on my laptop.

Eevee, with one black frame right in the middle:
spoon-eevee0001-0120.webm

Cycles, everything is fine:
spoon-cycles0001-0120.webm

**System Information** Operating system: Darwin-19.3.0-x86_64-i386-64bit 64 Bits Graphics card: AMD Radeon Pro 555X OpenGL Engine ATI Technologies Inc. 4.1 ATI-3.5.5 **Blender Version** Broken: version: 2.82 (sub 7), branch: master, commit date: 2020-02-12 16:20, hash: `77d23b0bd7` A 3D object with a reflective texture turns black when flattened by scaling x0 along an axis ~ I was rendering a transform of a reflective hemisphere from concave to convex, seeing how the reflection changes over time. - With Cycles, this works as expected. Concave is warped one way, flat is a flat mirror, convex is warped the other way. - With Eevee, the concave and convex reflections are fine, but for one frame where the hemisphere has become a 2D circle, it shows up black. Example file attached, saved at the half-way point in the animation where this bug occurs: [spoon.blend](https://archive.blender.org/developer/F8380159/spoon.blend) Videos attached to show the effect just in case this is hardware-specific and only happens on my laptop. Eevee, with one black frame right in the middle: [spoon-eevee0001-0120.webm](https://archive.blender.org/developer/F8380166/spoon-eevee0001-0120.webm) Cycles, everything is fine: [spoon-cycles0001-0120.webm](https://archive.blender.org/developer/F8380167/spoon-cycles0001-0120.webm)

Shish commented

2020-02-29 15:32:30 +01:00

Added subscriber: @shish_blender

blender-admin commented

2020-02-29 15:32:30 +01:00

#97031 was marked as duplicate of this issue

Jeroen Bakker commented

2020-03-02 08:39:35 +01:00

Changed status from 'Needs Triage' to: 'Confirmed'

Rodrigo commented

2020-04-15 04:14:34 +02:00

Added subscriber: @Rodrigo_S.T

Clément Foucault changed title from ~~Black reflections in eevee~~ to Zero sized normals if the object scale to zero in one axis.

2020-04-23 18:06:28 +02:00

Clément Foucault commented

2020-04-23 18:12:17 +02:00

Added subscribers: @brecht, @fclem

Clément Foucault commented

2020-04-23 18:12:17 +02:00

This happens because the object is scaled to 0.0 in one axis.

This also happens for workbench too.

@brecht what does cycles do to prevent this?

This happens because the object is scaled to 0.0 in one axis. This also happens for workbench too. @brecht what does cycles do to prevent this?

Brecht Van Lommel commented

2020-04-23 19:50:35 +02:00

It might be this code to invert degenerate matrices:

  if (UNLIKELY(!transform_matrix4_gj_inverse(R, M))) {
    /* matrix is degenerate (e.g. 0 scale on some axis), ideally we should
     * never be in this situation, but try to invert it anyway with tweak */
    M[0][0] += 1e-8f;
    M[1][1] += 1e-8f;
    M[2][2] += 1e-8f;

    if (UNLIKELY(!transform_matrix4_gj_inverse(R, M))) {

I doubt Cycles handles scale 0 well in all cases though, in general such transforms with zero scale along one axis should not be expected to give meaningful results.

It might be this code to invert degenerate matrices: ``` if (UNLIKELY(!transform_matrix4_gj_inverse(R, M))) { /* matrix is degenerate (e.g. 0 scale on some axis), ideally we should * never be in this situation, but try to invert it anyway with tweak */ M[0][0] += 1e-8f; M[1][1] += 1e-8f; M[2][2] += 1e-8f; if (UNLIKELY(!transform_matrix4_gj_inverse(R, M))) { ``` I doubt Cycles handles scale 0 well in all cases though, in general such transforms with zero scale along one axis should not be expected to give meaningful results.

Clément Foucault commented

2020-04-27 20:41:28 +02:00

Well we use the matrices computed by the depsgraph so that would have to be fixed here.

But since this matrix can be used for other purpose, I would not touch it for the sake of fixing this corner case.
If we were to recompute the inverse matrix again with this fix just for display purpose, we would take a significant performance hit.

Well we use the matrices computed by the depsgraph so that would have to be fixed here. But since this matrix can be used for other purpose, I would not touch it for the sake of fixing this corner case. If we were to recompute the inverse matrix again with this fix just for display purpose, we would take a significant performance hit.

Omar Emara commented

2022-04-04 15:44:27 +02:00

Added subscriber: @Leo-Luther

Philipp Oeser removed the

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Zero sized normals if the object scale to zero in one axis. #74317