Good clod hierarchy efficiency metric #1026
Replies: 1 comment 1 reply
-
|
The metrics the demo outputs per-level are the ones that were used during development too. Note that these are computed per level and there's some expected curve for how they change depending on the mesh structure, and the optimum for the metrics also somewhat depends on the mesh in some cases - for example, while the number of connected components per cluster should generally start at 1 and slowly creep up, if the source mesh has a lot of disjointed pieces, that growth will accelerate regardless of the underlying algorithm. The metrics also naturally depend on the configuration settings, such as the group size. So that has to be taken into consideration when doing any comparisons; you don't really need to vary the group size per scene too much, but I've seen some folks use suboptimally small numbers such as 2-4 cluster groups, which is not enough for the hierarchical simplification to work with. As far as the full-tree metrics, probably the most useful metric is the cut size; the demo computes a cut from a given viewpoint if instructed via env. variables, but you can also easily compute a number of cuts from increasing distances. For well-formed graphs you'd expect a reasonably smooth curve where both cluster and triangle counts fall off with distance of each cut as a power function. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Hi, Arseny!
I was working on developing a clod system myself and noticed that there are a lot of important decisions to be made. All of them heavily affect the resulting hierarchy. Then I decided to look for established solutions and wanted to compare them. The comparison however proved to be challenging.
Would you mind sharing some criteria/metrics you look for when working on the demo?
Beta Was this translation helpful? Give feedback.
All reactions