# Spheres are not perfectly round

 01 January 2013 #1 Broodstar Veteran   portfolio Michael Rauwerdink Mr The Animation School Cape Town, South Africa   Join Date: Jan 2012 Posts: 53 Spheres are not perfectly round Something I have noticed in Softimage and Maya. Spheres for some reason, even if the subdivisions are equal, are not round, but in fact oval. Why is this? Must I change the topology to fix this? Even when I bump the base resolution to 20u and 20v subdivisions, it still is oval, most noticeable when you rotate along the z-axis (you notice a slight bobble in the rotation) Any solutions to this? I would have thought it would be a common problem... share quote
 01 January 2013 #2 cojam Banned portfolio chris varible, Afghanistan   Join Date: Nov 2012 Posts: 177 All poly spheres are an approximation, no matter what program, the higher in u and v you go the closer the approximation to a sphere. Try your test with a Nurbs sphere, and give it too a high u,v, you'll see that as it mathematically defined it approximates the sphere shape better at lower u and v numbers. share quote
 01 January 2013 #3 Broodstar Veteran   portfolio Michael Rauwerdink Mr The Animation School Cape Town, South Africa   Join Date: Jan 2012 Posts: 53 Why does a poly sphere exist then? Surely if a cube can retain its shape when you press "3" to preview smooth it, then a sphere should be able to do it too? Edit: So I did some research, both on the internet and some in Maya (yes, I know this is the softimage section, but the results are the same) and here is what I came up with: As you can see NONE of the methods are perfectly round at a low resolution. And yes, it has to be a polygon. The Dodecahedron and Icosahedron might appear to be round, but they actually have slight dents and imperfections in them that are quite noticeable. What other methods are there? Last edited by Broodstar : 01 January 2013 at 04:12 PM. share quote
 01 January 2013 #4 grahamef Expert Grahame Fuller Tech Writer Autodesk M&E Canada   Join Date: Apr 2005 Posts: 415 The vertices all do lie on the mathematical sphere. The problem is not the sphere, but comes from the Catmull-Clark subdivision used to smooth it -- the poles pull more strongly on the shape because they have many more edges connected. You'll get closer to a sphere if you start with an octahedron, increase the Geodesic Frequency, and then smooth it. But it still won't be perfect -- you simply can't do a perfect sphere using polygons, subdees, or even cubic NURBS. It's good enough for animation though, which is the point. share quote
 01 January 2013 #5 Broodstar Veteran   portfolio Michael Rauwerdink Mr The Animation School Cape Town, South Africa   Join Date: Jan 2012 Posts: 53 How would I model/animate a robot with ball-joints then? How do people in industry model spheres? One might think this is a stupid thing to ask, but as you can see from my picture it is actually quite a big issue. share quote
 01 January 2013 #6 grahamef Expert Grahame Fuller Tech Writer Autodesk M&E Canada   Join Date: Apr 2005 Posts: 415 You only need to get close enough for the purpose of your animation. You can use higher-resolution geometry where you need a better approximation, and you can also hide problem areas away from the camera. share quote
01 January 2013   #7
Quote:
 Originally Posted by grahamef You only need to get close enough for the purpose of your animation. You can use higher-resolution geometry where you need a better approximation, and you can also hide problem areas away from the camera.

I was just goin to say the same thing. If you have a ball joint for a robot, your best bet is to just use the standard mesh sphere and increase the U and V segments...may 60+ for both? If you don't need to use smoothing (+,-) key then don't, just stick with the base mesh with a higher segment count for your case.

Now, lets say your ball joint has some detail in it, like grooves and such. Depending on what you need you can always do that with the higher segment base mesh, which could be problematic depending on what you need to do since the mesh is more dense, or you could do the main modeling in a lower mesh resolution and then use the (+,-) to set your smoothing - this however, could lead again to the main issue you are having as not being perfectly sphere. If you have to do it this way you have a couple of options; you could make two spheres: One as your reference (a higher segemented sphere as that I mentioned earlier), and the other is the sphere you modeled on. Then, just adjust and match your modeled sphere that has smoothing applied to it, to the reference sphere (High-poly base mesh) as best that you can and you should be good to go.

The other option, which can work on a modeled sphere, is to instead highlight the edges and use the loop cut function, and click the check box to tell the newly created curves to follow the base mesh curvature approximation...Alternatively, I also find highlighting the polygons you want smoothed and then go to Local Subdivison in the menu gives more options and somewhat a different result then just using the (+,-) solely, which could help you out.

Sorry about the longer post, but I just wanted to give you some options.

01 January 2013   #8
Quote:
 Originally Posted by Broodstar How would I model/animate a robot with ball-joints then? How do people in industry model spheres? One might think this is a stupid thing to ask, but as you can see from my picture it is actually quite a big issue.

I can assure you that in 16 years doing this for a living, most of which spent in the feature film industry, spheres not being spherical enough has NEVER been an issue.

Some topologies piling too many poles and pissing PRMan off or singularities at such poles in shading might have been, but nothing without known solutions, but even at extreme close ups on spherical objects (and I'm talking close up like an eye being bigger than the screen) a 16 rows sphere with C-C SDS has always been plenty precise.

It simply isn't a big issue, it's in fact completely a non-issue.

It's not a Soft or Maya issue either as much as one of the geometry descriptors and interpolators common in the creative industry (C-C or rarely D-S SDS, low degree NURBS etc.), but it's never really presented a problem, as the socket will be approximated the same way, and the human eye is terrible at picking up the discrepancies we're talking about.
__________________
Come, Join the Cult http://www.cultofrig.com - Rigging from First Principles

 01 January 2013 #9 Broodstar Veteran   portfolio Michael Rauwerdink Mr The Animation School Cape Town, South Africa   Join Date: Jan 2012 Posts: 53 What are the known solutions you speak of? My main concern and reason that I want it to be perfectly round and why I mentioned the robots joints is because when you rotate an imperfect sphere, you will notice a bobble, and my eye and I am sure anyone's eye will pick that up. Notice the distance at the top when I rotate your so called perfect 16 row sphere, clearly visible bobble when animating. The bobble will always be there, no matter how high the resolution of the sphere, because of the way it gets approximated, which is retarded I must say. So far no one has told me how they make their spheres or how industry makes their spheres. The only solution I have found so far is to make a nurbs sphere and convert into polygon, perfect sphere BUT has 39 rows, which is quite high poly (if my goal is to have it as low poly as possible) share quote
 01 January 2013 #10 Cinnsealach PRO   portfolio Andy Kinsella Manchester, United Kingdom   Join Date: Jan 2005 Posts: 191 Maybe you could scale the control mesh slightly in Y or XZ? share quote
 01 January 2013 #11 Broodstar Veteran   portfolio Michael Rauwerdink Mr The Animation School Cape Town, South Africa   Join Date: Jan 2012 Posts: 53 I could, but I cannot perfectly scale it. Also, why should that be necessary? When you click "Create Sphere" it should be perfect from the start. share quote
 01 January 2013 #12 cojam Banned portfolio chris varible, Afghanistan   Join Date: Nov 2012 Posts: 177 How high did you go? try a sphere u,v = 512. Then rot in Z, see any wobble? share quote
 01 January 2013 #13 Broodstar Veteran   portfolio Michael Rauwerdink Mr The Animation School Cape Town, South Africa   Join Date: Jan 2012 Posts: 53 I am trying to have it low poly though...512 is just insane. share quote
01 January 2013   #14
I shared with you my examples/methods on how to do it - they work/worked for me when I did robot animations...so, I dont know beyond that. I just whiped up a sample scene myself in Softimage and I have no visable issue - 60/60 for the sphere (see images). You could always scale it if you need to get it just right, which was mentioned previously...typing in a numerical value in the SRT boxes can give you a more precise scaling than just by using the mouse. All-in-all, you need a greater base (higher-poly) starting mesh to have the smoothing be less effective in its distortion when applied.

If you want the base mesh to be low-poly, then you will have to scale it as suggested. Although, how many polys are you really saving by using a lower poly mesh and then smoothing it, versus using a higher poly mesh without smoothing...

As Jaco mentioned earlier, the eye is not going to pick up the issue you are talking about, and if you have to zoom in on certain parts, then smooth it if you need too, the extra computing power will not be that much.
Attached Images
 Sphere_1.jpg (83.6 KB, 9 views) Sphere_2.jpg (97.2 KB, 7 views) Sphere_3.jpg (86.1 KB, 8 views)

01 January 2013   #15
Quote:
 Originally Posted by Broodstar What are the known solutions you speak of?

Re-read the post, I was referring to issues coming from topological singularities that relate to surfacing/rendering Not to shapes.

Quote:
 My main concern and reason that I want it to be perfectly round and why I mentioned the robots joints is because when you rotate an imperfect sphere, you will notice a bobble, and my eye and I am sure anyone's eye will pick that up.

Sure, you pick it up in a still fo the viewports, chances you will in a moving and moblurred shots are 0.

Quote:
 The bobble will always be there, no matter how high the resolution of the sphere, because of the way it gets approximated, which is retarded I must say.

At a high enough geometrical resolution it will NOT be there.

Quote:
 So far no one has told me how they make their spheres or how industry makes their spheres.

We have, the way you find so flawed is how we do it, and do so without problems I'd add.

Quote:
 The only solution I have found so far is to make a nurbs sphere and convert into polygon, perfect sphere BUT has 39 rows, which is quite high poly (if my goal is to have it as low poly as possible)

If you get as close as you show to those items, why would you want it low poly? No point to it whatsoever since it's a rigid object and you're well away still from polys being denser than your fragment level.

Honestly, mate, you are making a huge deal of something that people "in the industry" simply find a non issue.
You're wasting time and getting stuck on something you should have left behind you a while ago, but you seem intent in seeking someone validating your original position, no matter what gets posted.
__________________
Come, Join the Cult http://www.cultofrig.com - Rigging from First Principles

 Posting Rules You may not post new threads You may not post replies You may not post attachments You may not edit your posts vB code is On Smilies are On [IMG] code is On HTML code is Off CGSociety Society of Digital Artists www.cgsociety.org Powered by vBulletinCopyright ©2000 - 2006, Jelsoft Enterprises Ltd.