Laplacian Deform incorrect with > 10K faces #39004

New Issue

George Anderson · 2014-03-06T23:21:12+01:00

George Anderson commented

2014-03-06 23:21:12 +01:00

System Information
OSX 10.8.5

Blender Version
2.70RC1

Exact steps for others to reproduce the error

Add cube
Subdivide all faces 6 times (using subdivide tool, not the modifier) [there should be 24576 faces]
Select several verts, and add them to a vert group
Add a Laplacian Deform modifer, with that vert group
Bind

On my computer, the mesh distorts like crazy. If I subdivide 4 or fewer times, the results are perfectly correct. If I subdivide 5 times, then it "jumps" after binding, though not as severely as with 6 times.

It appears to want to pull the mesh in to the origin, while still respecting the added constraints.

**System Information** OSX 10.8.5 **Blender Version** 2.70RC1 **Exact steps for others to reproduce the error** 1) Add cube 2) Subdivide all faces **6 times** (using subdivide tool, not the modifier) [there should be 24576 faces] 3) Select several verts, and add them to a vert group 4) Add a Laplacian Deform modifer, with that vert group 5) Bind On my computer, the mesh distorts like crazy. If I subdivide **4 or fewer times**, the results are perfectly correct. If I subdivide **5 times**, then it "jumps" after binding, though not as severely as with 6 times. It appears to want to pull the mesh in to the origin, while still respecting the added constraints.

George Anderson commented

2014-03-06 23:21:12 +01:00

Changed status to: 'Open'

George Anderson commented

2014-03-06 23:21:12 +01:00

Added subscriber: @supergra

Patrice Bertrand commented

2014-03-07 16:46:26 +01:00

Added subscriber: @PatriceBertrand

Bastien Montagne commented

2014-03-18 16:08:29 +01:00

Added subscriber: @mont29

Alexander Pinzon was assigned by Sergey Sharybin

2014-03-21 10:47:02 +01:00

Sergey Sharybin commented

2014-03-21 10:47:02 +01:00

Added subscribers: @AlexanderPinzon, @Sergey

Sergey Sharybin commented

2014-03-21 10:47:02 +01:00

@AlexanderPinzon, ind checking on this one?

Alexander Pinzon commented

2014-03-21 13:32:46 +01:00

I'll check.

Brecht Van Lommel commented

2014-09-20 20:19:54 +02:00

Added subscriber: @brecht

Brecht Van Lommel commented

2014-09-20 20:19:54 +02:00

This is a precision issue in OpenNL / SuperLU. I've seen similar issues with UV unwrapping of high poly meshes.

Doing s/float/double for all files in intern/opennl solves the problem here, and seems to help nicely with UV unwrapping too. Run time is almost the same, doubles are pretty fast on modern CPUs and might help the solver to converge faster too. The main issue is doubling of memory usage, but I guess that's acceptable if the solver is otherwise quite useless for high poly meshes anyway.

This is a precision issue in OpenNL / SuperLU. I've seen similar issues with UV unwrapping of high poly meshes. Doing `s/float/double` for all files in `intern/opennl` solves the problem here, and seems to help nicely with UV unwrapping too. Run time is almost the same, doubles are pretty fast on modern CPUs and might help the solver to converge faster too. The main issue is doubling of memory usage, but I guess that's acceptable if the solver is otherwise quite useless for high poly meshes anyway.

George Anderson commented

2014-09-20 20:41:54 +02:00

Single precision bites again...That's been a perpetual headache for using (otherwise stellar) Blender my work.

Nice find.

I don't suppose there's a special build option to universally convert Blender to using doubles is there? That level of precision would be worth the memory overhead for me (not for 99% users, I know).

Single precision bites again...That's been a perpetual headache for using (otherwise stellar) Blender my work. Nice find. I don't suppose there's a special build option to universally convert Blender to using doubles is there? That level of precision would be worth the memory overhead for me (not for 99% users, I know).

Brecht Van Lommel commented

2014-09-20 20:45:40 +02:00

There exists no build option for it, and I don't think it would be easy to add. Globally replacing all floats by doubles can't be done automatically, there's various places where things would break (file read/write especially).

blender-admin commented

2014-09-26 00:04:10 +02:00

This issue was referenced by 0b12e61040

This issue was referenced by 0b12e6104007bb74c1f1450c5a4ee75e81cdfbdc

Brecht Van Lommel commented

2014-09-26 00:08:33 +02:00

Changed status from 'Open' to: 'Resolved'

Brecht Van Lommel closed this issue

2014-09-26 00:08:33 +02:00

Brecht Van Lommel commented

2014-09-26 00:08:33 +02:00

Closed by commit 0b12e61040.

Closed by commit 0b12e61040.

Brecht Van Lommel commented

2014-09-26 00:10:18 +02:00

Fix committed now, thanks for the report.

Here's an example of how this helps for UV unwrap, before and after.

Fix committed now, thanks for the report. Here's an example of how this helps for UV unwrap, before and after. ![uv_unwrap_opennl_precision.png](https://archive.blender.org/developer/F112542/uv_unwrap_opennl_precision.png)

blender-admin commented

2014-10-03 15:09:25 +02:00

This issue was referenced by ac234067e1

This issue was referenced by ac234067e1025dbcf0e3e25ae6ab780536ce11f7

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Laplacian Deform incorrect with > 10K faces #39004