Polycount? Understanding model efficiency.

Polycount

Not all models are created equal. With differences in size, complexity, accuracy, and use, 3D models can vary greatly. This variety can often make it difficult to compare models to one another, let alone determine whether a model is efficient enough for a given application. With the increase in demand for 3D models in virtual reality, augmented reality, web and mobile-based applications, it is critical to understand how a model’s polycount helps determine its efficiency and applicability.

-Subdivision of common polygon shapes into tris.

What is a Poly?

Poly is short for polygon. A polygon is a face that is created by connecting three or more line segments into a closed-form, like a square, triangle, or hexagon. In 3D models, the geometry is usually broken down into the smallest form of polygon, the triangle. Every 3D model can be divided into triangles or tris, no matter how complex or straightforward. Even rectangles in many 3D applications are split diagonally to form two tris. In this way, a cube would have a poly count of 6 squares or 12 tris. Even geometry that appears round or circular is really just segmented into small faces. The smaller the faces, the smoother the appearance yet, the higher the polycount.

-Cylinder models with varying levels of polycount. Low to highpoly from right to left.

Why Does the Poly Count Matter?

The polycount of a model, or how many polygons a model has, is one of the best ways to measure the efficiency of a model. You can determine a model’s efficiency by comparing its visual quality to its polycount. Typically, the higher the visual quality and the lower the polycount, the more efficient the model. This is often referred to as lowpoly or a model with a low polycount. While it is easy to make a lowpoly model, it is considerably harder to make one that also looks good. Simply put, it is a balance between visual quality and polycount that makes a good lowpoly model.  

When efficiency is not a priority, polycount can also be a good way to determine the quality or detail level of a model. In this case, a model with a higher polycount will typically be the more detailed model. This can be important for objects that will be viewed up close or for applications that require accuracy and high detail. Keep in mind that the visual quality and the polycount are separate aspects of a model and are not interchangeable. Just because a model has a high polycount does not mean it will look good. You should always evaluate both the polycount and the visual quality of a model together.

When a model is in an application that requires it to be downloaded from the internet or rendered in real-time, like a mobile or web-based app, the efficiency of the model can be more important than its visual appearance. This is because the model is loaded and rendered live, in real-time, for the program. Higher polycount models have more information and therefore take longer to load, as well as longer to render.  

Why is this important? Performance, plain and simple. Load speed and smooth performance are deal breakers in mobile and web-based applications. Most people aren’t willing to wait more than 5 or 10 seconds for a web page or game to load before they lose interest and sub-par performance in a VR game can cause the game to appear choppy which is enough to make you sick, literally.

-Varying levels of polycount and visual quality with a stylized lowpoly model on the left and a progression of low to highpoly models to the right. Note the difference in appearance.  

Polycount becomes increasingly important for a model that you wish to duplicate multiple times around a scene, like trees, rocks, or grass. If you have a scene that has a forest with 1000 copies of the same tree in it, reducing that model by 1000 polys will result in a scene that has a million fewer polygons.

How “good” Does it need to be?

Visual quality and efficiency are often opposing ends of a spectrum for 3D models. The higher the visual quality, the lower the efficiency, and vice versa. As performance is often vital, you need to determine the level of visual quality a model needs as much as possible to achieve the desired performance in your final application. Not every model is viewed in the same way. Some models will be viewed up close, while others may only be seen from very far away. The distance at which a model will be viewed should factor into your requirements of visual quality, or how good the model looks. A model that is only ever viewed at a distance can have a much lower visual quality because the viewer will not be able to notice a substantial difference. Therefore, the model can have many fewer polys without worrying about the loss of visual quality. This can allow for high performance, highly efficient models because you lowered the visual quality criteria.  

Using textures or images to replace more complex geometry is a great way for a lowpoly model to have a visual quality similar to a highpoly version. For instance, you may use an image of a wall with windows instead of having the windows modeled or an image of a branch with leaves instead of modeling each twig and leaf. This gives the illusion of a highly detailed model without the high polycount. In some instances, you can even get away with using a flat 2d object or sprite with a texture to replace a more complicated 3d model. This video shows how this concept was used to great effect in Mario Kart 64. It should also be noted that the number and size of the textures in a model also have a big impact on a model’s efficiency and can lead to extremely poor performance, so finding an appropriate balance is key.

-Diagram showing the use of textures(images) to replace complex geometry. The model on the left uses images with windows and doors instead of geometry.

Know Your Polycount

Understanding the context of how a 3D model will be used allows us to find or create the right type of model for our specific use. It allows us to single out where high-quality models are needed and where they aren’t, as well as when and to what extent the efficiency of a model is important. Polycount is really just a metric. We at Concept3D can use this metric to help evaluate the applicability of a model, but only if we have a complete understanding of how that model will be used.