Wrong direction with recalculate normals #43848

New Issue

Mikhail Rachinskiy · 2015-03-01T12:44:19+01:00

Mikhail Rachinskiy commented

2015-03-01 12:44:19 +01:00

System Information
Windows 8.1 x64
Renderer: GeForce GTX 860M/PCIe/SSE2

Blender Version
Broken: 2.73.9 9f6f151

Short description of error
Normals recalculated inside with certain geometry.

Exact steps for others to reproduce the error

Open blend file: normals_outside.blend
Press Ctrl N (see how normals turned inside on the left model)

The difference between models:

**System Information** Windows 8.1 x64 Renderer: GeForce GTX 860M/PCIe/SSE2 **Blender Version** Broken: 2.73.9 9f6f151 **Short description of error** Normals recalculated inside with certain geometry. **Exact steps for others to reproduce the error** - Open blend file: [normals_outside.blend](https://archive.blender.org/developer/F146508/normals_outside.blend) - Press Ctrl N (see how normals turned inside on the left model) The difference between models: ![diff.png](https://archive.blender.org/developer/F146506/diff.png)

Mikhail Rachinskiy commented

2015-03-01 12:44:19 +01:00

Changed status to: 'Open'

Mikhail Rachinskiy commented

2015-03-01 12:44:19 +01:00

Added subscriber: @MikhailRachinskiy

Julian Eisel commented

2015-03-01 14:14:23 +01:00

Added subscriber: @JulianEisel

Julian Eisel commented

2015-03-01 14:14:23 +01:00

There is an option in the Redo Panel called "Inside" which should help you getting the desired result.
Does that help?

There is an option in the Redo Panel called "Inside" which should help you getting the desired result. ![recalculate_normals_bug.png](https://archive.blender.org/developer/F146526/recalculate_normals_bug.png) Does that help?

Mikhail Rachinskiy commented

2015-03-01 15:16:53 +01:00

No, it does not help, because then normals on the right model will turn inside.

Bastien Montagne commented

2015-03-01 16:10:45 +01:00

Added subscriber: @mont29

Bastien Montagne commented

2015-03-01 16:10:45 +01:00

Not sure this is actually a bug… At least, it behaves accordingly to what you may expect from our code.

Thing is, we choose a first face, and if its normal is not 'aligned' with the vector from an average center point to that face's center, we flip it. Then, all other faces are aligned to that reference one.

Now, currently, we choose the farthest face as ref one, in that case the tip-top one, and you can see that the line between that face's center and the average center of the mesh is nearly orthogonal to its normal…

There are two things that annoy me in our code, though:

Why choose the farthest face? Do not see any particular reason (pro or con) for that…
Our average center point does not takes into account faces' surface, which means a set of small faces tend to 'drag' that center point to them, facilitating extreme cases like in this example. Tried taking surfaces' area into account and did not change anything in this case though.

Campbell, would like your feedback on this one?

Not sure this is actually a bug… At least, it behaves accordingly to what you may expect from our code. Thing is, we choose a first face, and if its normal is not 'aligned' with the vector from an average center point to that face's center, we flip it. Then, all other faces are aligned to that reference one. Now, currently, we choose the farthest face as ref one, in that case the tip-top one, and you can see that the line between that face's center and the average center of the mesh is nearly orthogonal to its normal… There are two things that annoy me in our code, though: * Why choose the farthest face? Do not see any particular reason (pro or con) for that… * Our average center point does not takes into account faces' surface, which means a set of small faces tend to 'drag' that center point to them, facilitating extreme cases like in this example. Tried taking surfaces' area into account and did not change anything in this case though. Campbell, would like your feedback on this one?

Bastien Montagne commented

2015-03-01 16:10:57 +01:00

Added subscriber: @ideasman42

Julian Eisel commented

2015-03-01 18:14:10 +01:00

@MikhailRachinskiy, even though I agree with Bastien that this could/should be improved, I think you can get the desired result by doing the normal correction on each mesh island separately, so you select one by mouse hovering it and pressing L, correct normals, then the same with the other one. You can then use the "Inside" option for the one that doesn't look correct. For me this seems to work fine.

Bastien Montagne commented

2015-03-01 18:28:08 +01:00

@JulianEisel this is not the issue here. The question is, that algo is supposed to recompute normals outside, and in this case it fails.

This is of course not a dramatic bug (if we even consider this a bug), it’s easy to work around it, but there’s no reason to tag this as incomplete, issue is easily confirmed.

@JulianEisel this is not the issue here. The question is, that algo is supposed to recompute normals **outside**, and in this case it fails. This is of course not a dramatic bug (if we even consider this a bug), it’s easy to work around it, but there’s no reason to tag this as incomplete, issue is easily confirmed.

Mikhail Rachinskiy commented

2015-03-01 18:43:06 +01:00

@JulianEisel I have an addon which turns non-manifold geometry in to manifold and then execute boolean operation, this is where I use "Make Normals Consisten" operator, so in my case I cannot manually check each object for bad normals.

blender-admin commented

2015-03-02 05:51:07 +01:00

This issue was referenced by 6373078321

This issue was referenced by 63730783218cc2c7f21fe4547beecd470397f35c

Campbell Barton commented

2015-03-02 05:57:19 +01:00

Changed status from 'Open' to: 'Resolved'

Campbell Barton closed this issue

2015-03-02 05:57:19 +01:00

Campbell Barton commented

2015-03-02 05:57:19 +01:00

Closed by commit 6373078321.

Closed by commit 6373078321.

Campbell Barton commented

2015-11-02 09:32:39 +01:00

Note, This solution failed in some cases.

Committed alternate fix 92ab3ba385

Note, This solution failed in some cases. Committed alternate fix 92ab3ba385

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

Wrong direction with recalculate normals #43848