Mesh is not being deformated correctly via Mesh Deform Modifier (quads, tris behave better) #91777

New Issue

Arthur Shapiro · 2021-09-28T14:17:38+02:00

Arthur Shapiro commented

2021-09-28 14:17:38 +02:00

System Information
Operating system: Windows-10-10.0.19041-SP0 64 Bits
Graphics card: NVIDIA GeForce GTX 1050/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 471.41

Blender Version
Broken: version: 2.93.4, branch: master, commit date: 2021-08-31 09:23, hash: b7205031ce
Worked: (newest version of Blender that worked as expected)

Short description of error
I have an Object 1 that I use as a Mesh Deformer target to on Object 2. Both are meshes, of course. Object 1 is parented to an Armature with just 1 Bone.
When I rotate the bone, the Object 2 is deformed incorrectly. It looks like "waves" are going through it.

Exact steps for others to reproduce the error
Bind Object 2 to and Object 1 with Mesh Deform Modifier. I use Precision of 6 which I think must be enough for such a simple construction. Then Rotate the Main Bone.
I also added the Camera parented to the Bone, so that you could use view from camera to see changes, relative to the rotation movement itself.

Note: triangulating the cage mesh with a modifier during bind seems to eliminate most of the wiggeling.

Mesh_Deform_Bug.blend

Mesh_Deform_Bug.mp4

**System Information** Operating system: Windows-10-10.0.19041-SP0 64 Bits Graphics card: NVIDIA GeForce GTX 1050/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 471.41 **Blender Version** Broken: version: 2.93.4, branch: master, commit date: 2021-08-31 09:23, hash: `b7205031ce` Worked: (newest version of Blender that worked as expected) **Short description of error** I have an Object 1 that I use as a Mesh Deformer target to on Object 2. Both are meshes, of course. Object 1 is parented to an Armature with just 1 Bone. When I rotate the bone, the Object 2 is deformed incorrectly. It looks like "waves" are going through it. **Exact steps for others to reproduce the error** Bind Object 2 to and Object 1 with Mesh Deform Modifier. I use Precision of 6 which I think must be enough for such a simple construction. Then Rotate the Main Bone. I also added the Camera parented to the Bone, so that you could use view from camera to see changes, relative to the rotation movement itself. Note: triangulating the cage mesh with a modifier during bind seems to eliminate most of the wiggeling. [Mesh_Deform_Bug.blend](https://archive.blender.org/developer/F10650284/Mesh_Deform_Bug.blend) [Mesh_Deform_Bug.mp4](https://archive.blender.org/developer/F10650323/Mesh_Deform_Bug.mp4)

Arthur Shapiro commented

2021-09-28 14:17:38 +02:00

Added subscriber: @zickkie

Arthur Shapiro commented

2021-09-28 15:14:25 +02:00

I am sorry but I am a bit emotional on this issue. There are multiple 3D projects and dozens of 3D animators in our studio - and all this made with Blender. We really love it and appreciate its functionality.
And then we create a project with characters, which rigs made with Mesh Deform - and what I get? Totally unstable and non-viable midifier that forces me to spend hundred of hours to rebuild all the project pipeline from the beginning. That is totally disappointing. Especially when I inderstand that this big - I am sure!- didn't appeared yesterday. That is just me who accidentally faced it.
Why do we ever need the tool that is broken from the beginning?

I am sorry but I am a bit emotional on this issue. There are multiple 3D projects and dozens of 3D animators in our studio - and all this made with Blender. We really love it and appreciate its functionality. And then we create a project with characters, which rigs made with Mesh Deform - and what I get? Totally unstable and non-viable midifier that forces me to spend hundred of hours to rebuild all the project pipeline from the beginning. That is totally disappointing. Especially when I inderstand that this big - I am sure!- didn't appeared yesterday. That is just me who accidentally faced it. Why do we ever need the tool that is broken from the beginning?

Alexander Gavrilov commented

2021-09-28 15:33:16 +02:00

Added subscriber: @angavrilov

Alexander Gavrilov commented

2021-09-28 15:33:16 +02:00

Maybe I'll have a look later - it could be something as simple as a floating point precision issue. Could be something difficult as well of course, so no guarantees.

Arthur Shapiro commented

2021-09-28 15:34:01 +02:00

In #91777#1227380, @angavrilov wrote:
Maybe I'll have a look later - it could be something as simple as a floating point precision issue. Could be something difficult as well of course, so no guarantees.

Thanks in advance!

> In #91777#1227380, @angavrilov wrote: > Maybe I'll have a look later - it could be something as simple as a floating point precision issue. Could be something difficult as well of course, so no guarantees. Thanks in advance!

Philipp Oeser commented

2021-09-29 11:29:57 +02:00

Added subscribers: @brecht, @lichtwerk

Philipp Oeser commented

2021-09-29 11:29:57 +02:00

Changed status from 'Needs Triage' to: 'Needs Developer To Reproduce'

Philipp Oeser commented

2021-09-29 11:29:57 +02:00

Can at least confirm the behavior (have not checked further).

Note though that it is somewhat noted as a possible limitation here https://docs.blender.org/manual/en/dev/modeling/modifiers/deform/mesh_deform.html

Significant changes to the entire cage mesh (such as rotating the cage upside down) can cause noticeable artifacts.

@brecht might also know more?

Can at least confirm the behavior (have not checked further). Note though that it is somewhat noted as a possible limitation here https://docs.blender.org/manual/en/dev/modeling/modifiers/deform/mesh_deform.html > Significant changes to the entire cage mesh (such as rotating the cage upside down) can cause noticeable artifacts. @brecht might also know more?

Alexander Gavrilov commented

2021-09-29 11:38:24 +02:00

Added subscriber: @jpbouza-4

Alexander Gavrilov commented

2021-09-29 11:38:24 +02:00

Further investigation yesterday in #blender-riggers with hints from @jpbouza-4 revealed this is fixed by triangulating the mesh with a modifier during bind, so it seems the binding code does something funky with quads - possibly it's too sensitive to non-planar quads or something.

So while I think this still needs investigation, at least there seems to be a workaround now.

Further investigation yesterday in #blender-riggers with hints from @jpbouza-4 revealed this is fixed by triangulating the mesh with a modifier during bind, so it seems the binding code does something funky with quads - possibly it's too sensitive to non-planar quads or something. So while I think this still needs investigation, at least there seems to be a workaround now.

Philipp Oeser commented

2021-09-29 12:09:27 +02:00

In #91777#1227913, @angavrilov wrote:
Further investigation yesterday in #blender-riggers with hints from @jpbouza-4 revealed this is fixed by triangulating the mesh with a modifier during bind, so it seems the binding code does something funky with quads - possibly it's too sensitive to non-planar quads or something.

So while I think this still needs investigation, at least there seems to be a workaround now.

Wow, good to know!

> In #91777#1227913, @angavrilov wrote: > Further investigation yesterday in #blender-riggers with hints from @jpbouza-4 revealed this is fixed by triangulating the mesh with a modifier during bind, so it seems the binding code does something funky with quads - possibly it's too sensitive to non-planar quads or something. > > So while I think this still needs investigation, at least there seems to be a workaround now. Wow, good to know!

Philipp Oeser changed title from ~~Mesh is not being deformated correctly via Mesh Deform Modifier~~ to Mesh is not being deformated correctly via Mesh Deform Modifier (quads, tris behave better)

2021-09-29 12:11:06 +02:00

Brecht Van Lommel commented

2021-09-29 18:09:26 +02:00

Changed status from 'Needs Developer To Reproduce' to: 'Needs User Info'

Brecht Van Lommel commented

2021-09-29 18:09:26 +02:00

The algorithm is known to not be rotation invariant, I don't think there's a way to eliminate the wiggling without using a very different implementation. It's a documented limitation and not a bug by itself, but if there's ways to improve accuracy that's interesting.

The binding process already triangulates the mesh internally, but that triangulation may be different, or there may be a bug in the implementation.

However I saw no improvement triangulating the cage mesh in this example .blend file.

The algorithm is known to not be rotation invariant, I don't think there's a way to eliminate the wiggling without using a very different implementation. It's a documented limitation and not a bug by itself, but if there's ways to improve accuracy that's interesting. The binding process already triangulates the mesh internally, but that triangulation may be different, or there may be a bug in the implementation. However I saw no improvement triangulating the cage mesh in this example .blend file.

Arthur Shapiro commented

2021-09-29 22:26:29 +02:00

@brecht In my file I do these steps:

deleted existing MDef Modifier
add Triangulate Modifier to MDef Cage
Bind at 6 level of Precision
Delete Triangulate Modifier
... and that produces very accurate result.
Mesh_Deform_Bug_Triangulated.blend
MD_Triangulation.mp4

@brecht In my file I do these steps: - deleted existing MDef Modifier - add Triangulate Modifier to MDef Cage - Bind at 6 level of Precision - Delete Triangulate Modifier ... and that produces very accurate result. [Mesh_Deform_Bug_Triangulated.blend](https://archive.blender.org/developer/F10677920/Mesh_Deform_Bug_Triangulated.blend) [MD_Triangulation.mp4](https://archive.blender.org/developer/F10677932/MD_Triangulation.mp4)

Juan Pablo Bouza commented

2021-09-30 00:14:13 +02:00

I have tested this with Arthur and can say that triangulating before binding removes almost all of the distortion when rotating.

In this video, the first model is bound with quads and it wobbles a lot. The second model was bound with a triangulated cage and the result is almost perfect. At the end you can see the comparison between the wires of both meshes.

I guess that the triangulation @brecht and @angavrilov have mentioned is not occurring at the binding process, otherwise there shouldn't be any difference in the result.

wobbly_mdef.webm

I have tested this with Arthur and can say that triangulating before binding removes almost all of the distortion when rotating. In this video, the first model is bound with quads and it wobbles a lot. The second model was bound with a triangulated cage and the result is almost perfect. At the end you can see the comparison between the wires of both meshes. I guess that the triangulation @brecht and @angavrilov have mentioned is not occurring at the binding process, otherwise there shouldn't be any difference in the result. [wobbly_mdef.webm](https://archive.blender.org/developer/F10679738/wobbly_mdef.webm)

Brecht Van Lommel commented

2021-09-30 16:18:41 +02:00

Added subscriber: @ideasman42

Brecht Van Lommel commented

2021-09-30 16:18:41 +02:00

@zickkie you also ended up turning off the Dynamic option in the video, which makes a difference. But without that indeed I can see triangulation helps.

This problem may have been introduced by 368bd3457. @ideasman42, do you remember if that was to fix a specific bug?

I can see how a discrepancy between the intersected coordinate in a triangle and the interpolated coordinate in non-planar polygons would cause problems. It's not clear to me what the problem is described in that commit.

@zickkie you also ended up turning off the Dynamic option in the video, which makes a difference. But without that indeed I can see triangulation helps. This problem may have been introduced by 368bd3457. @ideasman42, do you remember if that was to fix a specific bug? I can see how a discrepancy between the intersected coordinate in a triangle and the interpolated coordinate in non-planar polygons would cause problems. It's not clear to me what the problem is described in that commit.

Brecht Van Lommel commented

2021-09-30 16:19:00 +02:00

Changed status from 'Needs User Info' to: 'Confirmed'

Philipp Oeser removed the

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Mesh is not being deformated correctly via Mesh Deform Modifier (quads, tris behave better) #91777