Efficient use of polygons
This tutorial is aimed at newcomers to modelling. It partially discusses a problem that I call “square modeling”. This is where a significant number of the polygons in a model are (nearly) perfect squares. I’m not taking about quads – they have 4 sides and 4 corners, but this doesn’t mean they are square.
Use it or lose it
Often when you start modeling you end up using more polygons than you need to define a shape, and you keep a lot of square edges. The most common mistakes that newcomers make is to create another row of edges (known as a loop) just to define a corner on one area of a mesh. The other new vertices and egdes that were created are not used to define the shape of the model, so in effect they are being wasted.
I must stress that sometimes leaving these extra polygons is a good idea – either for better use of textures when UV mapping, or for better lighting results. However, in most cases this extra geometry can be removed, or used to define more shape.
If its there, use it.
The tutorial
This is a very short and simple tutorial showing how I often see shape definition being done inefficiently, and describes a more elegant method. It deals SOLELY with shaping without excess geometry.
This tutorial is also aimed SOLELY at static parts of a model, not areas with joints that should deform. Joints are a case when keeping or adding geometry is very useful.
The pipe
Firstly, lets get a simple shape with some angles on it. I’ve drawn a very simple pipe that has a slight kink in it, and a then surrounded it with a cube to block out the shape.
|
 |
| The initial sketch. | The initial sketch with a box surrounding the extents. |
Method one
Now that we have the basic box, lets make the shape.
Firstly, here is the way I often see it being done – the modeller looks at each angle individually, without considering the shape as a whole. Since they consider each angle individually, they construct each angle individually.
 |
 |
| The modeler creates a new row of edges for every angle on the object. There are 2 angles at each side (2 sides shown), making 4 new rows of edges. | Once the loops are created the vertices are pulled into shape. |
The model looks correct, and works perfectly. However, there are more polygons there than need to be. Only part of the geometry is used to define the overall shape, and some parts simply do nothing – the look like big squares. This is “square modeling”.
If you look at bottom left section of the second image above, you can see 4 vertices in a perfect vertical line. The middle 2 are serving no purpose. These are wasted. Where edges meet at 180 degrees, the connecting vert is often unnecessary – either give it a purpose of defining some shape, or simply remove it.
Method two
Lets have a look at another method to build the same shape, this time considering the shape as a whole, looking at volumes of mass.
 |
 |
| I create a new loop for every PAIR OF ADJACENT angles on the object. There are 2 angles at each side, so that is 2 new rows of edges, not 4 as before. | Once the new rows of edges are created the vertices are pulled into shape. |
The first thing to note is that the model also looks correct – it is exactly the same shape as before. However, there are less polygons than the first method. How many? Lets work it out.
We’ll have to make some assumptions:
- The pipe has no ends
- The pipe has 4 sides
- 1 quad = 2 polygons
The maths part
In the first method, there were 5 rows, or 5 loops of quads. The pipe has 4 sides.
4 sides * 5 sections of quads * 2 (to convert quads to triangles) = 40 triangles/polygons.
In the second method, there were 3 sections, or 3 loops of quads.
4 sides * 3 sections of quads * 2 (to convert quads to triangles) = 24 triangles/polygons.
Method 2 uses almost HALF the number of polygons that method 1 uses.
This was a very simple tutorial based on a simple feature – a pipe – yet the same method works perfectly well on characters. With characters it can provoke a more organic flow.
Points to note
As mentioned at the start, this tutorial is aimed SOLELY at static parts of model, not areas with joints that should deform. When you have a join, such as an elbow or a knee, you need to have those extra rows (loops). If you don’t have them, the model will collapse horribly during animation.
With some models, adding the extra geometry used in the first method will make the model much easier to UVmap, and may improve lighting – squares provide a better generic surface to calculate light and shadow over as they are based on 2 identical triangles.