Reinventing the wheel
polycounter lvl 15
Instead of using a standard truncated cone/tapered cylinder/frustum, I've been rotating the top Ngon a bit, because I feel it looks a bit better. I was wondering what the community thought of this.
I realize this isn't the most interesting of shapes, nor does this show massive visual differences. But I feel that this variation of making a truncated cone/tapered cylinder warrants some thought because:
-it costs no extra polies
-i think it improves visuals
-it's technically more consistent
-it's a pretty basic form, and can be used in a lot of models, for instance as bolts or buttons on machinery, as nubs on off-road wheels, and as ocular for cameras and rifle scopes.
1 - Silhouette
The standard method varies between 8-10 sides, depending on viewing angle, which causes 'bulges' to form at the top left and right side and make the shape a bit wobbly.
My method maintains it amount of sides, volume and shape better and has no/less blobbyness.
2 - Enhanced silhouette
The standard method has the sides slide in and out of sight in a rather jumpy fashion. The the sides coming into view are also looking a bit boxy. This happens because when a new side comes into view, 2 planarly aligned triangles come into view, once every X time.
My method has the triangles coming into view more consistently, about one tri every 0.5X time. This leads to the appearance and disappearance looking smoother. It also makes the 'shaded' area taper out more consistent.
3 - Wireframe model
The standard method has quads which in most engines would be triangulated into either [/] or[\] which gives some inconsistencies. These triangles are also irregular (3 different line lenghts/corner angles)
My method is triangulated by default, and thus uses the same model in every engine. Triangles are isosceles triangles, which means they have at least 2 sides of equal length. (with some models you'll even have equilateral triangles) So the model is a bit more consistent. It also has more evenly sized triangles (less long thing tris) but I highly doubt this will give a noticeable performance improvement.
4 - Vertex-Shading
The standard method has banded highlights, resulting in a nice horizontal gradient. But the somewhat parallel banding also results in the cylinder obviously having 8 facets instead of being perfectly round.
My method has slightly more spread out highlights, and because the vertices don't line up there's no horizontal gradient. This makes it so that the amount of sides of the cylinder is less visible. Unfortunately it makes it look a bit dented.
The shaded model is arguably better or worse when using my method, depending on use (intensity if highlights, new/old looking asset, etc) but it's also the least important aspect, seeing as vertex shading is mostly negated by applying a normalmap.
The only absolute downsides about my method are:
-it costs you an extra minute
-you might have to sacrifice a small amount of texture space (because the unwrap would look like /________/ instead of |________|)
TL;DR:
Left or right model, which is better?
I realize this isn't the most interesting of shapes, nor does this show massive visual differences. But I feel that this variation of making a truncated cone/tapered cylinder warrants some thought because:
-it costs no extra polies
-i think it improves visuals
-it's technically more consistent
-it's a pretty basic form, and can be used in a lot of models, for instance as bolts or buttons on machinery, as nubs on off-road wheels, and as ocular for cameras and rifle scopes.
1 - Silhouette
The standard method varies between 8-10 sides, depending on viewing angle, which causes 'bulges' to form at the top left and right side and make the shape a bit wobbly.
My method maintains it amount of sides, volume and shape better and has no/less blobbyness.
2 - Enhanced silhouette
The standard method has the sides slide in and out of sight in a rather jumpy fashion. The the sides coming into view are also looking a bit boxy. This happens because when a new side comes into view, 2 planarly aligned triangles come into view, once every X time.
My method has the triangles coming into view more consistently, about one tri every 0.5X time. This leads to the appearance and disappearance looking smoother. It also makes the 'shaded' area taper out more consistent.
3 - Wireframe model
The standard method has quads which in most engines would be triangulated into either [/] or[\] which gives some inconsistencies. These triangles are also irregular (3 different line lenghts/corner angles)
My method is triangulated by default, and thus uses the same model in every engine. Triangles are isosceles triangles, which means they have at least 2 sides of equal length. (with some models you'll even have equilateral triangles) So the model is a bit more consistent. It also has more evenly sized triangles (less long thing tris) but I highly doubt this will give a noticeable performance improvement.
4 - Vertex-Shading
The standard method has banded highlights, resulting in a nice horizontal gradient. But the somewhat parallel banding also results in the cylinder obviously having 8 facets instead of being perfectly round.
My method has slightly more spread out highlights, and because the vertices don't line up there's no horizontal gradient. This makes it so that the amount of sides of the cylinder is less visible. Unfortunately it makes it look a bit dented.
The shaded model is arguably better or worse when using my method, depending on use (intensity if highlights, new/old looking asset, etc) but it's also the least important aspect, seeing as vertex shading is mostly negated by applying a normalmap.
The only absolute downsides about my method are:
-it costs you an extra minute
-you might have to sacrifice a small amount of texture space (because the unwrap would look like /________/ instead of |________|)
TL;DR:
Left or right model, which is better?
Replies
Are you talking about wavyness in the normal map? Because the right one would also give issues like these (polycount wiki).
EarthQuake:
Agreed, I stubmled upon something something similar a while ago
Wailingmonkey:
Cool, we/I should experiment with variations of these.
would look on lowpoly. Just for experimentin's sake.
I definitely collapse edges on spheres like eq showed though.
http://wiki.polycount.net/Sphere_Topology
Why not bother? If something can improve performance/visuals I'm all for it. All depending on the amount of effort vs improvement ofcourse.
Aniceto:
well obviously that's going to improve how things look. But I'm talking about circumstances where you can't do that (without big impact) such as rivets on a plane, where it would cost thousands of polies to add 1 extra side. Or panels with buttons. Or possibly very lowpoly games.
PredatorGSR:
I realise it's not a big issue, nothing revolutionary, but I was seeing if it was possible to improve without addinf more edges.
EricChadwick:
Not at all. Honoured, actually.
So I've spent some time in Max messing about with domes, here's the results:
-the octa based ones are rather sucky, with weird shading
-50▲ icosa based one is nicely balanced, for when you need less sides (it's pretty much the EQ topology, but 5 sides instead of six)
-64▲ sphere based is like EQ's + chamfered nipple for a pinch better side silhouette (allowing a higher dome)
if people are going to take anything away after reading this thread it should be the basic ideas about enhancing a low poly mesh, not specifically how to make an 8 sided cylinder look slightly better from a certain angle.
poopinmymouth showed examples of optimised spheres years ago, so there's hardly any reinventing of wheels going on in here...unless its an 8 sided wheel
By rotating the edges with some intersected geometry, we can round out the 'glass' a bit at no extra polygon-cost. This is quite a specific situation, but it can also be useful to round out some buttons or screws on some objects.
There is actually more texel distortion on 'straight' cones than there is on a 'turned' cone. Here are two nearly identical models and textures, the only difference are:
-left one has the top border rotated by 18 degrees (360 degrees, 10sides, rotated half a 'side')
-it's textures have been adjusted accordingly.
You'll notice that the texels on the 'straight' cone are distorted in size, but that all pixels are also slanted to one side, like parallelograms. All the texels on the bottom left triangle of each poly slant to the left, while those on the upper right triangle slant to the right.
The 'turned' cone shows the same amount of size/width distortion, but has a lot less slanting. Only the texels that are actually intersected by an edge show slanting. The other ~85% is nice and rectangular.