Mesh Smooth Node #86903

New Issue

Hans Goudey · 2021-03-24T20:36:56+01:00

Hans Goudey commented

2021-03-24 20:36:56 +01:00

This task is for a port of the "Smooth" modifier to a node.

Details

Mesh smoothing can be thought of as a function of smoothing an arbitrary attribute.
- This simply requires the use of position vector smoothing.
The field evaluation system allows to implement smoothing for different domains. This implies that smoothing will be used for different domains based on the connections of this domain. Extraneous domain interpolations are excluded, no adaptation is required.
- Domain must be implicitly defined by the context.
- Geometry, for calculating connections, can also be obtained from the context.
All input values can be fields.
- For qualitative values, everything is simple.
- Quantitative ones require an auxiliary structure (Count of iteration).

Implementation Questions
The factor is clamped between 0 and 1. This is a consequence of the mixer design used for the attributes. Can be corrected by simply refusing that the factor always accumulates from positive components D16351: BKE_attribute_math: DefaultMixer: remove asserts for negative weight.

This task is for a port of the "Smooth" modifier to a node. ![image.png](https://archive.blender.org/developer/F13808089/image.png) **Details** - Mesh smoothing can be thought of as a function of smoothing an arbitrary attribute. - This simply requires the use of position vector smoothing. - The field evaluation system allows to implement smoothing for different domains. This implies that smoothing will be used for different domains based on the connections of this domain. Extraneous domain interpolations are excluded, no adaptation is required. - Domain must be implicitly defined by the context. - Geometry, for calculating connections, can also be obtained from the context. - All input values can be fields. - For qualitative values, everything is simple. - Quantitative ones require an auxiliary structure (Count of iteration). **Implementation Questions** The factor is clamped between 0 and 1. This is a consequence of the mixer design used for the attributes. Can be corrected by simply refusing that the factor always accumulates from positive components [D16351: BKE_attribute_math: DefaultMixer: remove asserts for negative weight](https://archive.blender.org/developer/D16351).

Hans Goudey commented

2021-03-24 20:36:56 +01:00

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

Hans Goudey commented

2021-03-24 20:36:56 +01:00

Added subscriber: @HooglyBoogly

Hans Goudey commented

2021-03-24 20:37:27 +01:00

Added subscriber: @howardt

Kenzie commented

2021-04-01 06:26:29 +02:00

Added subscriber: @KenzieMac130

Kenzie commented

2021-04-01 06:26:29 +02:00

Would it be possible to generalize this to other attributes than position? I could see users wanting to smooth things like uv's and normals/direction vectors as well. I'm wondering if this might be a good place to do an attribute smooth/blur.

Hans Goudey commented

2021-04-01 15:33:16 +02:00

Yeah, that's an interesting idea. I think it's definitely possible. The algorithm is specific to meshes anyway, since it uses edges, so I'm not sure I would make it an attribute node, but an input for an attribute choice could always be added in the future. Though I think it would need to use the point domain.

Howard Trickey commented

2021-04-04 17:51:39 +02:00

After talking with Hans about this some, I think I will generalize this node to have both an input attribute and result attribute. For now, both should be on the Point domain, though I can think of ways to generalize to other domains. On the point domain, the semantics are: the nominal smoothed result attribute value is gotten by taking, for each edge attached to the point, the average attribute value at each end of the edge, and averaging all those together. (This is what the Smooth modifier does for "position".) Then the Factor, which can be per-Point if the Factor input is also on the Point domain, is used as the interpolation factor between the original attribute value and the nominal smoothed result.

Also, upon reflection, a soft range of about -3 to 3 seems approximately right for Factor. There is some usefulness of going outside the range of strictly between "original" and "nominal smoothed value", but it starts to get crazy (at least for position) when the factor is more than about 3.

After talking with Hans about this some, I think I will generalize this node to have both an input attribute and result attribute. For now, both should be on the Point domain, though I can think of ways to generalize to other domains. On the point domain, the semantics are: the nominal smoothed result attribute value is gotten by taking, for each edge attached to the point, the average attribute value at each end of the edge, and averaging all those together. (This is what the Smooth modifier does for "position".) Then the Factor, which can be per-Point if the Factor input is also on the Point domain, is used as the interpolation factor between the original attribute value and the nominal smoothed result. Also, upon reflection, a soft range of about -3 to 3 seems approximately right for Factor. There is some usefulness of going outside the range of strictly between "original" and "nominal smoothed value", but it starts to get crazy (at least for position) when the factor is more than about 3.

Gregory commented

2021-04-04 19:34:14 +02:00

Added subscriber: @GregoryS

Himanshi Kalra commented

2021-04-09 20:40:05 +02:00

Added subscriber: @Calra

Howard Trickey commented

2021-04-11 17:08:31 +02:00

I've made an initial try at this, in D10951.

I've made an initial try at this, in [D10951](https://archive.blender.org/developer/D10951).

Zach Eastin commented

2021-11-03 16:24:36 +01:00

Added subscriber: @Zeastin

Iliya Katushenock was assigned by Hans Goudey

2022-08-05 17:36:41 +02:00

Paul Larson commented

2022-09-10 09:11:16 +02:00

Added subscriber: @GeorgiaPacific

blender-admin commented

2022-12-07 18:22:44 +01:00

This issue was referenced by d68c47ff34

This issue was referenced by d68c47ff347bbb3824c27387f3387bd0583158a2

Hallam Roberts commented

2022-12-13 16:30:55 +01:00

Added subscriber: @MysteryPancake

Hallam Roberts commented

2022-12-13 16:30:55 +01:00

Is there likely to be an "Attribute Sharpen" node in the future? If so, it might be better to generalize this node into a "Filter" node (similar to the compositor) or a "Kernel" node

Iliya Katushenock commented

2022-12-13 16:37:20 +01:00

So far, no plans, the design of the work will be interesting

Hans Goudey commented

2022-12-13 16:37:45 +01:00

We've never talked about a "Sharpen" node. Maybe it would be useful, not sure though. In that case, I'm not sure there would be a benefit compared to having two separate nodes, unless maybe the kernel itself was an input socket.

Hallam Roberts commented

2022-12-13 17:03:55 +01:00

Ah nevermind, it is possible to sharpen by using a negative weight. It is cool this is possible :)

Ah nevermind, it is possible to sharpen by using a negative weight. It is cool this is possible :) ![image.png](https://archive.blender.org/developer/F14055771/image.png) ![image.png](https://archive.blender.org/developer/F14055782/image.png)

Iliya Katushenock commented

2022-12-13 17:04:56 +01:00

Added subscriber: @JacquesLucke

AgentA1cr commented

2023-01-13 15:29:52 +01:00

Added subscriber: @AgentA1cr

AgentA1cr commented

2023-01-13 15:29:52 +01:00

It would be quite useful to be able to set the weight of individual neighbor elements. Is that something that might be implemented in the future?

Jacques Lucke commented

2023-01-13 17:24:47 +01:00

Yes, might be implemented in the future. We need some more code infrastructure to support that well, but there is no fundamental reason not to support it in the future.