jacobeverist / opaque Goto Github PK

When computing splices for each possible branch point of a shoot on a parent, sometimes the child skeleton will move and extend the parent at its terminals. however, this change is not reflected when computing the splices and is returning a shortened splice. Subsequent fitting and evaluation of pose posture curves will result in a poor fit and the branch point will have a low probability leading to its exclusion.

The pathological case results in the control point pushing the child shoot farther and farther away from the parent terminals and towards egregious elphantine shoot appendages.

I've noticed that the angle of the junction seems to be in the wrong direction sometimes. Specifically, what should be around 0 degrees in a Cross-Junction, turns out to be around pi/-pi. However, the for downward portion of the junction, the angle is correctly described as -pi/2.

When finding a path between points on the global skeleton, it sometimes fails because the start and target node are the same. The big question is why is the same node for the start and finish being selected, should we check for this case or let it continue to error out until we find the root cause.

In cases where two paired posture curves fit on to separate splices, the batchLocalizeParticles operation will only evaluate fitness for both of them as a team for a single splice. So that if they are best fit on separate splices, this case is not evaluated properly.

However, the two nodes will still be shoot classified properly. Just the localization procedure will be screwed up.

Arc distances as a parameter to determine the location of child's branch point from a parent are useful. However, if the distance from the parent's origin to the branch point changes, then the arc distance will change also. This can happen if the parent is also branching from a grandparent and the branch point is changed.

That or the parent's origin gets changed by backtracking to that location and continuing mapping.

Plotted terminal point on all_branch_plot_*.png show the terminal points in red, but the plotted splices don't always extend to these points. This could be that the skeleton doesn't extend that far or that it is not correctly selecting the terminal as a starting point.

Inverted branches occur when the child shoot branches inward onto the parent. This occurs when there is confusion on which end of the child is branching out of the parent. This seems to occur with the static path.

This may be a general symptom or just a vestigial artifact of the static map that we will gradually phase out.

For the control point at certain configurations, the child skeleton will not overlap with the parent skeleton in any plausible way. However, this configuration is still given serious consideration when it should be filtered out early.

Requires a skeleton overlap metric for initial plausibility of the branch point so these situations can be filtered out early.