There are many local optima

< Previous | Contents | Manuals Home | Boris FX | Next >

‌There are many local optima

The solver is looking for the best explanation for your tracking data. You'd like for it to find the single best explanation (a global optimum) but that isn't mathematically possible.

It's like finding the top of a mountain. At the top, every surrounding point is downhill from the top. But the mountain has many boulders on top of it. Which one is the highest? And there are pebbles on top of the boulders, and grains of sand on top of the pebbles, and You can survey the top of any boulder/pebble/sand grain, but there are

infinitely many and it takes a while to measure each one, so you can't do them all.

Furthermore, those height measurements are what are subject to the accuracy limitations described above: they are the sum of many error values. So the height measurements have some error to them. And the pebbles on the top of a boulder are almost at the same height to start with, and the grains of sand on top of the pebbles are even closer together in height. You're comparing their height with tiny differences, the accuracy matters.

Since you probably want a result well before the end of the universe, SynthEyes (or anything else) can't find the true global optimum (top of the mountain). It finds a local optimum, the top of some boulder/pebble/sand grain—hopefully one of the highest.

< Previous | Contents | Manuals Home | Boris FX | Next >