Weightpainting with "x-mirror" and "auto normalize" fails for center bones #29832

New Issue

Tobias Oelgarte · 2012-01-08T19:10:15+01:00

Tobias Oelgarte commented

2012-01-08 19:10:15 +01:00

%%%While weightpainting with "auto normalize" enabled, the x-mirror code is buggy and that in all combinations (x-mirror, topology mirror, x-mirror + topology mirror). It computes the weights wrong or ignores them entirely. I appended an example file. Just start painting and you will see that it's not mirrored as expected. My current version of blender is r43173.

Typical examples are:

Painting on left vertex has no influence on right vertex.
Sometimes it does work (not in my test scenario), but then it doesn't keep the other vertex groups (auto-normalization, not the one painted on) mirrored.

This behavior gets worse and worse with any additional changes to the weightpainting. It's really annoying and currently impossible to use. :-(

Can't it simply calculate the normalized result for one side and then copy it to the vertex from the other side? With copying i mean that .L and .R groups will be switched before transfering the values. This is simple and would always work.

%%%

%%%While weightpainting with "auto normalize" enabled, the x-mirror code is buggy and that in all combinations (x-mirror, topology mirror, x-mirror + topology mirror). It computes the weights wrong or ignores them entirely. I appended an example file. Just start painting and you will see that it's not mirrored as expected. My current version of blender is r43173. Typical examples are: * Painting on left vertex has no influence on right vertex. * Sometimes it does work (not in my test scenario), but then it doesn't keep the other vertex groups (auto-normalization, not the one painted on) mirrored. This behavior gets worse and worse with any additional changes to the weightpainting. It's really annoying and currently impossible to use. :-( Can't it simply calculate the normalized result for one side and then copy it to the vertex from the other side? With copying i mean that .L and .R groups will be switched before transfering the values. This is simple and would always work. %%%

Tobias Oelgarte commented

2012-01-08 19:10:15 +01:00

Changed status to: 'Open'

Campbell Barton commented

2012-01-11 09:23:02 +01:00

%%%confirmed, looking into it.%%%

Tobias Oelgarte commented

2012-01-11 10:52:16 +01:00

%%%I appended a little bit more advanced scenario (added one left and right bone) for testing purposes.

If it works correctly you should be able to paint on "B.L" and "B.R" should show the same but mirrored result. But additionally the groups "A" and "B" should also be mirrored. The last time i painted on the left upper leg (leg-upper.L) i had to notice that the right upper leg (leg-upper.R) was mirrored correctly, but the centered group for the hips (hip) was only changed on the left side and not affected on the right side.

One more thing comes to my mind: If i paint on ".L" oder ".R" groups i noticed that the strength is at best 0.5, even so if set to 1.0. I assume that it is applied as 1.0 but then mixed with the other sides old value, resulting in the average of both groups/sides. I don't know if this is intentional, but from a user perspective the painted weight should be the reference. It sets the value and anything else has to adjust to it, while leaving the currently painted group unaffected.%%%

%%%I appended a little bit more advanced scenario (added one left and right bone) for testing purposes. If it works correctly you should be able to paint on "B.L" and "B.R" should show the same but mirrored result. But additionally the groups "A" and "B" should also be mirrored. The last time i painted on the left upper leg (leg-upper.L) i had to notice that the right upper leg (leg-upper.R) was mirrored correctly, but the centered group for the hips (hip) was only changed on the left side and not affected on the right side. One more thing comes to my mind: If i paint on ".L" oder ".R" groups i noticed that the strength is at best 0.5, even so if set to 1.0. I assume that it is applied as 1.0 but then mixed with the other sides old value, resulting in the average of both groups/sides. I don't know if this is intentional, but from a user perspective the painted weight should be the reference. It sets the value and anything else has to adjust to it, while leaving the currently painted group unaffected.%%%

Campbell Barton commented

2012-01-13 06:59:27 +01:00

%%%thanks for the second example, helpted with testing.
fixed r43342. commit log.

also fix 3 other bugs / inconsistencies

when mirror is enabled, center verts get themselves as the mirror vertex, wpaint code didn't check for this case, mirroring the vertex to it self (would work, but unneeded).
normalizing after painting would scale the active weight too, now keep the painted weight and adjust others when normalizing.
VP_ONLYVGROUP wasn't respected for mirror (mirror vertex group would always be added).

this changes normalize behavior - comment from code:

/* note on normalize - this used to be applied after painting and normalize all weights,

in some ways this is good because there is feedback where the more weights involved would
'risist' so you couldn't instantly zero out other weights by painting 1.0 on the active.
However this gave a problem since applying mirror, then normalize both verts
the resulting weight wont match on both sides.
If this 'resisting', slower normalize is nicer, we could call
do_weight_paint_normalize_all() and only use...
do_weight_paint_normalize_all_active() when normalizing the mirror vertex.
- campbell
  */

Setting as closed but I'd like to know if this makes mirror/normalize workflow work smoothly for you in more complicated cases.
if not will re-open.%%%

%%%thanks for the second example, helpted with testing. fixed r43342. commit log. --- also fix 3 other bugs / inconsistencies * when mirror is enabled, center verts get themselves as the mirror vertex, wpaint code didn't check for this case, mirroring the vertex to it self (would work, but unneeded). * normalizing after painting would scale the active weight too, now keep the painted weight and adjust others when normalizing. * VP_ONLYVGROUP wasn't respected for mirror (mirror vertex group would always be added). --- this changes normalize behavior - comment from code: /* note on normalize - this used to be applied after painting and normalize all weights, * in some ways this is good because there is feedback where the more weights involved would * 'risist' so you couldn't instantly zero out other weights by painting 1.0 on the active. * * However this gave a problem since applying mirror, then normalize both verts * the resulting weight wont match on both sides. * * If this 'resisting', slower normalize is nicer, we could call * do_weight_paint_normalize_all() and only use... * do_weight_paint_normalize_all_active() when normalizing the mirror vertex. * - campbell */ Setting as closed but I'd like to know if this makes mirror/normalize workflow work smoothly for you in more complicated cases. if not will re-open.%%%

Campbell Barton commented

2012-01-13 07:01:46 +01:00

Changed status from 'Open' to: 'Resolved'

Campbell Barton closed this issue

2012-01-13 07:01:46 +01:00

Tobias Oelgarte commented

2012-01-13 14:10:48 +01:00

%%%I will test it out. Since i have to paint on a stomache with lots of overlapping weight groups (even left bones have little influence on right side, etc.) . That was usually the case in which the weightpainting didn't work as expected. So i will be compiling and do a lot of testing and come back if i find things that aren't as they should be.%%%

Tobias Oelgarte commented

2012-01-13 16:57:52 +01:00

%%%I found something. If a vertex lies at a center position (x=0.0) and you paint on a .L/.R vertex group to which this vertex not already belongs, then you will need to click/draw twice before a change is visible. Must be something like: First click assigns vertex to vertex group(s). Second click draws the actual weight to this groups.%%%

Tobias Oelgarte commented

2012-01-13 17:36:57 +01:00

%%%This seams also to affect the blur-brush on any part of the mesh. If you just hit on a vertex (neigbour of a vertex beloning to the group) it has no effect. As soon as a vertex that belongs to the group is included in the selection (brush area) it will be luring inside this region. But i don't know how the blur-brush is supposed to be working in weightpaint mode. Does it blur the vertices under the brush (average) or does it blur neighbouring vertices (like smooth in edit-mode)?%%%

Campbell Barton commented

2012-01-14 19:54:49 +01:00

%%%fix L/R double click on center verts r43382.

re: Blur brush - it finds the average value from all points under the brush and paints from it - I don;t really like this method much but it works as intended.

%%%

%%%fix L/R double click on center verts r43382. re: Blur brush - it finds the average value from all points under the brush and paints from it - I don;t really like this method much but it works as intended. %%%

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

Weightpainting with "x-mirror" and "auto normalize" fails for center bones #29832