Variable Width Offsets #499
Replies: 10 comments 12 replies
-
|
Hi Marius.
Yes, Clipper can certainly help, but you'll still need to do a lot of work too, and make relatively minor modifications to Clipper.
|
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the pointers Angus. |
Beta Was this translation helpful? Give feedback.
-
|
The code in all 3 languages is pretty much the same, but the compiled C++ code is significantly faster than C# and Delphi. |
Beta Was this translation helpful? Give feedback.
-
|
@AngusJohnson, following your suggestion, we were able to make some good progress. Here are initial tests results: The resulting curve, however, has jagged edges: Do you have an idea why this is happening? @WileyNg worked on the code and could provide you with specific implementation details if necessary. |
Beta Was this translation helpful? Give feedback.
-
|
http://www.angusj.com/clipper2/Docs/Units/Clipper/Functions/SimplifyPaths.htm |
Beta Was this translation helpful? Give feedback.
-
|
Looks like jagged edge is a common phenomenon and it is recommended that the native offset function is to be followed by SimplifyPaths() method then. |
Beta Was this translation helpful? Give feedback.
-
|
Also, see some draft variable offset code here: |
Beta Was this translation helpful? Give feedback.
-
I'd like to have another look at this, so a smallish data sample would be helpful. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for taking the time to investigate it @AngusJohnson! Top view: Perspective zoom: Also, please notice the vertical transition with offset values of 13+ meters: |
Beta Was this translation helpful? Give feedback.
-
|
Edit: OK, I can see now if I make the SVG image 10x bigger that there are tiny imperfections in the path, just as you've indicated. |
Beta Was this translation helpful? Give feedback.
-
|
These values are in meters (we're landscape architects, remember? :) ). Quite often we'd be working zoomed in on a feature like this. Also, you're only considering the XY plane, things get even weirder when we look at the curve in 3 dimensions (see above). |
Beta Was this translation helpful? Give feedback.
-
|
Sorry. You were the totally unworthy target of some of my minor frustrations at others who've asked much less considered questions. |
Beta Was this translation helpful? Give feedback.
-
|
No hard feelings @AngusJohnson. I appreciate the time you keep pouring into this library. It is of genuine help to us! |
Beta Was this translation helpful? Give feedback.
-
😁. Try this ... in OffsetPoint(): |
Beta Was this translation helpful? Give feedback.
-
|
Seems to work much better on 2D curves: It broke the 3D offset, though. This might, however, be due to my very rough port of your c++ code to our current c# implementation, I'll investigate further as well: |
Beta Was this translation helpful? Give feedback.
-
|
I ran some more tests: Before: After: Before: After: It seems to have helped with the first case, but it still struggles with the second one. |
Beta Was this translation helpful? Give feedback.
-
Marius, could you please provide the data for this second case (plus the offsetting code). |
Beta Was this translation helpful? Give feedback.
-
|
Sure thing, here is the code snippet ported from your c++ sample: private void OffsetPoint(Group group, Path64 path,
int j, ref int k)
{
// Let A = change in angle where edges join
// A == 0: ie no change in angle (flat join)
// A == PI: edges 'spike'
// sin(A) < 0: right turning
// cos(A) < 0: change in angle is more than 90 degree
double sinA = InternalClipper.CrossProduct(_normals[j], _normals[k]);
double cosA = InternalClipper.DotProduct(_normals[j], _normals[k]);
if (sinA > 1.0) sinA = 1.0;
else if (sinA < -1.0) sinA = -1.0;
if (cosA > 0.99) // miter all joins (even concave ones) when angle < 8.1 deg.
DoMiter(group, path, j, k, cosA);
else if (cosA > -0.99 && (sinA * _group_delta < 0))
{
// is concave
group._outPath.Add(GetPerpendic(path[j], _normals[k]));
// this extra point is the only (simple) way to ensure that
// path reversals are fully cleaned with the trailing clipper
group._outPath.Add(path[j]); // (#405)
group._outPath.Add(GetPerpendic(path[j], _normals[j]));
}
else if (_joinType == JoinType.Miter)
{
// miter unless the angle is so acute the miter would exceeds ML
if (cosA > _tmpLimit - 1) DoMiter(group, path, j, k, cosA);
else DoSquare(group, path, j, k);
}
else if (_joinType == JoinType.Square)
DoSquare(group, path, j, k);
else
DoRound(group, path, j, k, Math.Atan2(sinA, cosA));
k = j;
}And this is the geometry: |
Beta Was this translation helpful? Give feedback.
-
What I was after was the code you're using to generate the offset. For example ... And with offset of 20, I'm not seeing any imperfections in the solution (using my amended
(If you just click on the image above, GitHub will make a mess of it. So you'll need to save it to file and then open it with your web browser.) |
Beta Was this translation helpful? Give feedback.
-
|
I just saw 037d47c, my branch is slightly behind master. I'll update to the newest source and switch to c++ to make sure we are on the same page. Having said that, the example you sent above still has some orphaned vertices on the right hand side of the curve: [EDIT] the issue is symmetric and can be seen on the left hand side as well: |
Beta Was this translation helpful? Give feedback.
-
YOU WEREN'T SUPPOSED TO NOTICE!! 😱😜😁. OK, here's the deep gos ... And if you really zoom in .... There are also the tiniest positive windings within the negative windings that aren't being clipped. These can be removed by deleting paths with virtually no area, or by calling SimplifyPath() using an appropriate epsilon. |
Beta Was this translation helpful? Give feedback.
-
|
@AngusJohnson, it turns out that you were right assuming I was missing a basic setting: 2023-05-04.15-40-31.mp4I pulled the repo, built it with the most recent version of c++ code and set up an interactive testing environment in Rhino. The precision value, which drives the scaling factor, was set too low for my geometry. Initial tests seem to confirm that setting it just one higher fixes the issue. [EDIT] I did, however, replace the OffsetPoint() code with the snippet provided in this thread. |
Beta Was this translation helpful? Give feedback.
-
@AngusJohnson, thanks a million for creating and maintaining this fantastic library!
I work with landscape architecture. A relatively common challenge in the design process is finding variable-width offsets of linear features. Real-life example:
The top curve is the one I care about as a designer, the bottom one is a result of projecting perpendicular rays onto the underlying terrain. All is fine, as long as the guiding curve is not concave and/or the target surface relatively close. Things can get messy, though:
2023-04-20.08-28-47.-.Trim.mp4
In our profession, we face these challenges all the time:
Do you have an idea how Clipper could help here?
Beta Was this translation helpful? Give feedback.
All reactions