Massive landscapes, who doesn't dream about them. In outdoor games landscapes are one of the most crucial parts of rendering. They easily take up 60% of the screen, more than enough reason to make them look good.
Part of recent events I’ve taken the landscape rendering trough an development iteration to enhance the visual quality. The key to enhancing visual quality is detail. Too much detail is bad, too little detail is equally as bad. The detail needs to be just right. Our consumption for detail has outgrown to the point where modern-age storage facilities such as the hard-disk or even ram/vram are just not feasible to store all the detail.
An aspect of landscape rendering are the textures and the transitions between textures. Traditionally people use ‘splatting’ or ‘alpha-maps’. These approaches are fine but another technique caught my eye. In essence you create fractal such as fbM (fractional Brownian motion) to displace the slope & altitude and use a lookup texture to find the parameters for the layer that corresponding with the slope for the specific altitude. Below is a sample transition. More details of this method can be found on http://www.gamedev.net/blog/73/entry-1692117-terrain-texturing-explained/.
Another aspect of landscape quality is the resolution or better said how small are your polygons. The rule of thumb is that when all vertices are aligned on a single plane they can be unified into a single primitive. This unification/subdivision is quite neat for adding details like the curvature of an hill. When you elevate the terrain the tool can analyze if patches needs to be subdivided or unified and that will give the basis for an editable terrain.
However as great as this subdivision/unification system is it doesn’t add detail to your terrain. It is of course possible to dynamically subdivide the polygons based on lod. Traditionally this was done on the CPU and the algorithm was called ‘ROAM’. But with the rise of DirectX 11 / Opengl 4.0 hardware new pipeline was introduced the Tesselation pipeline.
The tessellation pipeline allows you to subdivide triangles or quadrilles on the GPU in parallel. This lends to a few things less bandwidth consumed for geometry upload/download, geometry stays valid until its edited, and we get to have detail polygons. Thanks to these detail polygons we can now add more detail.
Two things you need to look out for: Tessellation has the capabilities to generate both vertices at the edges of the primitive and subdivide the surface area itself, the generated edge vertices are not shared. What this means is if you subdivide the edges of your primitive non-deterministic or displace generated vertices in a non-deterministic manner cracks will appear. One of the mistakes I made in my initial tessellation was correlate the wrong edges with their corresponding gl_TessLevelOuter[n] variable.
Part of recent events I’ve taken the landscape rendering trough an development iteration to enhance the visual quality. The key to enhancing visual quality is detail. Too much detail is bad, too little detail is equally as bad. The detail needs to be just right. Our consumption for detail has outgrown to the point where modern-age storage facilities such as the hard-disk or even ram/vram are just not feasible to store all the detail.
An aspect of landscape rendering are the textures and the transitions between textures. Traditionally people use ‘splatting’ or ‘alpha-maps’. These approaches are fine but another technique caught my eye. In essence you create fractal such as fbM (fractional Brownian motion) to displace the slope & altitude and use a lookup texture to find the parameters for the layer that corresponding with the slope for the specific altitude. Below is a sample transition. More details of this method can be found on http://www.gamedev.net/blog/73/entry-1692117-terrain-texturing-explained/.
Another aspect of landscape quality is the resolution or better said how small are your polygons. The rule of thumb is that when all vertices are aligned on a single plane they can be unified into a single primitive. This unification/subdivision is quite neat for adding details like the curvature of an hill. When you elevate the terrain the tool can analyze if patches needs to be subdivided or unified and that will give the basis for an editable terrain.
However as great as this subdivision/unification system is it doesn’t add detail to your terrain. It is of course possible to dynamically subdivide the polygons based on lod. Traditionally this was done on the CPU and the algorithm was called ‘ROAM’. But with the rise of DirectX 11 / Opengl 4.0 hardware new pipeline was introduced the Tesselation pipeline.
The tessellation pipeline allows you to subdivide triangles or quadrilles on the GPU in parallel. This lends to a few things less bandwidth consumed for geometry upload/download, geometry stays valid until its edited, and we get to have detail polygons. Thanks to these detail polygons we can now add more detail.
Two things you need to look out for: Tessellation has the capabilities to generate both vertices at the edges of the primitive and subdivide the surface area itself, the generated edge vertices are not shared. What this means is if you subdivide the edges of your primitive non-deterministic or displace generated vertices in a non-deterministic manner cracks will appear. One of the mistakes I made in my initial tessellation was correlate the wrong edges with their corresponding gl_TessLevelOuter[n] variable.
Comments
Post a Comment