Le 12/09/2016 à 16:53, clipka a écrit :
> Am 11.09.2016 um 20:54 schrieb clipka:
>> Work in progress.
> 
> ... because something I still have to figure out is what output is most
> useful:
> 
> (A) the raw blob potential,
> (B) the difference between blob potential and threshold,
> (C) the distance to the blob surface, or
> (D) yet some other metric.
> 
> 
> (A) would have the advantage that the sum of any such patterns would
> behave just like a combo of the underlying blobs.
> 
> (B) would have the advantage that the pattern would be less dependent on
> the blob's specific settings, always giving a zero value at the blob's
> surface.
> 
> (C) would be advantageous for obvious reasons.
> 
> (D) may provide yet unforeseen other advantages.
> 


Do you normalize or truncate ? (aka, how do you handle entry above 1.0 or below 0.0 in
various ***_map ?)

(C) looks simple but are you sure you can get the smallest distance ?
especially when the strength and radius are different between components, and there is
negative strength in the equation.
I would not expect (C). It seems easy but fails as soon as there is a negative
strength in the blob, something you cannot forbid.

(A) is the most invariant: you can change the threshold without worry.

(D) :
1. recently, the product of the two max was considered (for something totally
different: bevelled and junctions)
2. what about the Max() of contribution (instead of L_1 (sum), it would be another
traditional metric (L_infinity metric IIRC)
3. and of course, we could look at L_2 (sqrt(sum(square of contribution))), aka
Euclidian distance. Can be funny or interesting with negative contribution

Of course (B) can be applied to (A) and any (D), reversing the effect at the surface
of the blob.

Post a reply to this message