I am developing a very simple tool whose task is to smooth the captured weight as fast as possible.
Motivation:
I know that nodes like Smooth SOP or Delta Mush already exist for this purpose, but in some scenarios the result of their work was not quite right or not fast enough. The goal of this competition is to see if we can make a tool as efficient and simple as possible using OpenCL updated in H20.
Description:
The node has no parameters yet except for one number of iterations of weight smoothing. Therefore, it is quite easy to use and does not need to pre-populate attributes, this node already does that. The demo file already has examples of usage. In some cases this asset is already faster than Smooth SOP by a factor of at least 2, but so far, it's not for everyone.
TODO:
- Can we rewrite the solution using local buffer and memory for optimization?
- Improve optimization for cases with more bones, right now this solution is slower than Smooth SOP.
- Add possibility to smooth regions using masks, attributes, groups.