Differential D7283
Weld Modifier - Average vertex weights of welded vertices Authored by Cody Winchester (CodyWinch) on Mar 31 2020, 5:08 AM.
Details This patch aims to give the user the option to average the merged vertex weights with the weld modifier. The current functionality adds together all of the vertices weights and clamps 0-1. This will add the weights together and divide by the total # of vertices. Vertices not assigned to the existing vertex group will be treated as a weight of 0.0f The option is now a single boolean that reuses the existing flag.
Diff Detail
Event TimelineComment Actions It seems that vertex weights are already interpolated in the last master. Could you check? Comment Actions Seems like you are correct. It appears to now add all the weights of the weld verts together. This actually results in weights above 1.0 from my small testing. I will take a look at the new code and see if I can better implement averaging as an option. Thanks for notifying me. Comment Actions Modified for new master. In the new master the vertex weights are all added together and can result in weights greater than 1.0f. Now the patch averages based on all of the vertices welded. So a vert not assigned to the vertex group acts as a weight of 0.0f. Comment Actions Updated for new master. New version has the vertex weight mixing method as an option for the user to choose from. There are 3 modes:
Comment Actions Thanks for the patch.
Comment Actions
I have another version of the patch that adds the average weights functionality as a single boolean. This would do away with the third option of using the lowest weight. This would address your concerns about the use of an additional flag and making the modifier struct larger. It would reuse the existing flag and only add 1 option to it.
Comment Actions Changed to a single option boolean. I thought the current weight mixing was using the highest vertex weight, but I was wrong and it adds the weights together then clamps to 0-1. So I have simplified the added change to just an averaging option. Comment Actions Are there any situations where the current behavior of adding all the vertex weights together and then clamping is actually preferred? I'm not sure I like the idea of adding an option for behavior that most people expect anyway. (IMO the current behavior mentioned above of adding them up would not be expected) If the option is necessary for performance I'd rather have a choice between "Nothing" and "Expected Behavior" than a choice between "Unexpected Behavior" and "Expected Behavior". Comment Actions I would agree that averaging should be the default behavior. In this case I am not sure what the "Nothing" behavior would be. Comment Actions This is not the proper way to fix this issue. The root of the issue comes from calling CustomData_interp() with NULL weights (it is the only place in code we do that currently, I think). Instead of generating average values, that function will instead just add them. I cannot see any good reason for this behavior, so added D9114: Refactor CustomData interpolation code., which indeed fixes the problem here. | ||||||||||||