[ < ] [ > ]   [ << ] [ Up ] [ >> ]         [Top] [Contents] [Index] [ ? ] [ Search: ] About Deferred Rendering

What is Deferred Rendering?

Deferred shading is a rasterization based rendering method that separates the rendering of geometry from the process of lighting that geometry. In the traditional approach to rasterization (commonly referred to as forward shading) the rendering and lighting of geometry are done in tandem. As a result, using forward shading tends to limit the number of light sources that can affect a single object. However, with deferred shading the cost of adding a light to a scene is independent of the number of objects in a scene. This property allows for a scene to contain significantly more dynamic light sources before performance becomes an issue.

Implementing Deferred Rendering

From a high-level deferred shading works in two main passes. In the first pass all geometry in a scene is drawn to a set of buffers referred to collectively as the GBuffer (short for Geometry Buffer). During this pass the GBuffer is filled with the information needed to evaluate the lighting equation. This information includes as a minimum the surface normal, surface position (stored as depth and reconstructed when needed) and shininess. In the second pass the filled GBuffer is used to evaluate the lighting equation for each light in the scene. The contribution from each light is added to, what is commonly referred to as, the accumulation buffer. At the start of the second pass the accumulation buffer is initialized to all black. During the pass the accumulation buffer will be filled with lighting information from each light. By the end of the pass the accumulation buffer will contain enough information to compute the final image.

To compute the lighting equation for a light source we must determine what pixels in the final image are affected by that light. To do this we take advantage of the rasterization capabilities of the GPU by rendering light proxy geometry. This proxy geometry is a mesh that approximates the size, shape, and position of the light source we are considering. As an example, we might use a cone mesh to approximate a spot light or a sphere mesh to approximate a point light. Ideally we want the proxy geometries approximation be as tight as possible so we do not needlessly consider pixels that are not affected by the light. The light proxy geometry is rendered using a shader (see section Shaders) that reads data from the GBuffer, computes the lighting equation, and outputs the result to the accumulation buffer at each pixel.

Difference between Deferred Rendering Techniques

Crystal Space implements two different deferred rendering techniques. The first and more traditional one is deferred shading. In this technique the GBuffer contains all information that is needed to compute the final image is stored in a single pass over the geometry and there's only one accumulation buffer to which the results from each light are added. The second technique is deferred lighting which only uses a minimal GBuffer containing the information needed to accumulate the results of each light, but not enough information to compute the final image. Most noteably neither the diffuse nor the specular color are stored as they don't vary for a pixel so they are not needed to accumulate the irradiance. The cost of this small GBuffer is that two accumulation buffers are needed (one for diffuse, one for specular light) and that a second pass over the geometry after the accumulation is required in which all extra information (such as diffuse and specular color, ambient, etc.) is fetched and the final image is produced.

Considerations when Using Deferred Shading

Despite the benefits in lighting that deferred shading provides there are multiple issues that should be considered when choosing this rendering method:

[ < ] [ > ]   [ << ] [ Up ] [ >> ]         [Top] [Contents] [Index] [ ? ]

This document was generated using texi2html 1.76.