FitCrv goes bananas on that one

[osuire]

To obtain a single NURBS curve from a multi-curve, I often rebuild with a high number of control points (here : 1000) and then use FitCrv.
In this particular instance, FitCrv goes totally crazy.

The whackyness changes depending on how many CP the curve has.

I can post to original curve to McNeel only, sorry.

[Gijs (Gijs de Zwart)]

yes, pls send the file (you can either upload or send me in pm)

[osuire]

@Gijs

yes, pls send the file

Done

[Gijs (Gijs de Zwart)]

RH-82749 FitCrv goes haywire

[osuire]

Here’s one I can share.
FitCrv doesn’t even bother on this one : null output.

240626_PROG_Cant fit this.gh (12.5 KB)

Looking for a similar tool, but one that will not crap in my boots when I need it the most.
Any ideas ?

[René_Corella]

@osuire

Any ideas ?

Does pulling the control points to the original curve then interpolating them with a periodic curve serve as a ‘similar’ solution?

Also, silly but sometimes using SimplifyCurve helps if there’s a tiny discrepancy somewhere.

@osuire

I often rebuild with a high number of control points […] then use FitCrv.

I assume the high number is to ensure high fidelity so maybe not worth asking about using smaller number.

I played with it - not sure what the issue would be without testing your proprietary figure.

Some options here:
240626_PROG_Cant fit this.gh (29.9 KB)

The bottom example references an old divide-by-curvature method.

[Gijs (Gijs de Zwart)]

changing the value slightly makes it work. The question is though, is 456 control points a significant improvement?

[markz]

One of my ‘tricks’ from the past is offsetting the curve first, say, by 0.010 (as long as this is greater than document tolerances).

[osuire]

@Gijs

The question is though, is 456 control points a significant improvement?

It’s an improvement over 1000, the intended value.

[osuire]

@markz

One of my ‘tricks’ from the past is offsetting the curve first, say, by 0.010 (as long as this is greater than document tolerances).

Whatever works ! I might give it a try.
Thanks

[osuire]

Thanks, I’ll try that out tomorrow

[Gijs (Gijs de Zwart)]

@osuire

It’s an improvement over 1000, the intended value.

your sample showed 500(503)

[osuire]

Yes, but in the initial post, the super crazy result was based on a curve with 1000.
As you throttle the CP count, there are various values that produce odd results, but the one with 1000 was completely bonkers, as you can see in my original post.

Actually, this brings back an old request I had for a tool that would do the reverse action to “Convert” which transforms a Nurbs into a polyline.
As inputs, I think it would be great to have the Tolerance as usual, plus a threshold angle where the initial polyline would be split, resulting in multiple Nurbs curves.
But above all, it would need to be ROBUST.

[markz]

Moving closer to origin and offsetting works

[osuire]

@René_Corella

The bottom example references an old divide-by-curvature method .

That’s smart !
I just needed to jack the “Curvature factor” a bit for the effect to show up :