1. Stylized Environmental Rendering
Michael Lester

2. Background
Rendering of natural environments is very difficult for photorealistic real-time graphics
It is very expensive and the results are often unsatisfactory.
Apply Non-photorealistic rendering (NPR) techniques to render a natural environment.
This is traditionally a very difficult area of real-time graphics.
That's where Non-photorealistic rendering comes in
Or stylized rendering, I'll use the two interchangeably

3. Stylized Rendering
Generate images that appear don't adhere to real-world rules
Convey information without being precise and complete
Stylized Rendering solves this problem in two ways
The generated images (can) take less time to compute because they are not as complex, yet they still convey the same information
And they can often look ''better”
The human brain can easily pick out errors in a photorealistic image (since we see them every day)
But stylized images don't have this problem
For example, in the screenshot on the left, you can clearly recognize an ocean, even though most of the water is a simple, flat blue
On the right, you can tell that the torches are burning, but the flames are much more simple than realistic fire

4. The Goal Apply artistic techniques to interactive graphics
Traditionally animated film
NPR research papers
Solve rendering problems by using an NPR style
Use the GPU as much as possible
Applicable to Games
The goal of my research is (very broadly) to study the techniques that artists use in illustrating outdoor environments, and apply them to real-time computer graphics
More specifically I'm interested in grass or fields and how characters interact with them
This leads to two paths of research:
I'll need to study traditionally animated film to pick out their techniques
And then apply them by building on Research in the area of Non- Photorealistic Rendering
Another of my goals for this project is to use the GPU as much as possible
Much of the research that is applicable to this project does not take full advantage of modern hardware.
The NPR field blossomed before programmable shaders became available, So many techniques are CPU heavy or rely on the fixed-function pipeline, which means they are quite slow
I'd like my system to take full advantage of the GPU
This means using the programmable shaders to do most of the work
And behind it all is the idea that I'd like my work to be applicable to games
Fast, easy to implement, and inexpensive

5. How do artists solve the problem?
They have it easy
Fixed viewpoint allows for image- based effects
Use several “tricks” that can be adapted to interactive graphics
So how do Artists handle grassy environments?
Well, they've got it easy
Film and images have a single, fixed viewpoint.
This means that they can get away with a lot,
Such as using a single static texture for grass.
Games, however, have arbitrary viewpoints.
This is the main challenge in adapting animation techniques to games.
For example,
When games use a single, static texture for grass, it is very obvious because of the lack of parallax when the viewpoint changes.
However, many of their tricks are applicable to interactive graphics
It just might take some work

6. “Deep” Grass Grass is simply a plane
Artists create the illusion of depth
Only the silhouette is significant
One technique that I think is quite applicable to games, and that I am focusing on, is deep grass
In this image, the artist creates the illusion of depth by using the silhouette only
The grass is just a simple plane, but the jagged overlapping of the grass on the character's edges make him appear to be sinking into the field
This is very applicable to games because characters often interact with grass by walking through it.
Many stylized games simply use a static grass texture
But this technique could be implemented very efficiently and would be a significant improvement

7. Applying to Graphics Graftals
Based on a paper by Markosian and Kowalski
A small texture element attached to an object
Represented with by a small geometric “fin”
Affected by viewing angle, distance, geometric collision, etc.
Abstracts the scene complexity away from the geometry
So how do we implement this for Real-Time Graphics?
We can use a concept known as Graftals
Originally proposed in two papers by Kowalski and Markosian
They are basically small, specifically textured geometric fins that are placed all around an object
They differ from normal geometry because they are decoupled from the modeled scene geometry
They are not always drawn
And they're not always drawn the same way
For example, graftals are commonly drawn only near the edges of objects
Or where to objects intersect, such as a character moving through a field
This has the affect that the Rendering engine now controls scene complexity, instead of the geometric models.
This gives the engine a lot of power
It can keep the level of detail consistant across the scene
Procedural animation is much easier
Which is great for wind effects
Control the rendering cost easily
It can be reduced if the system is heavily taxed

9. The System Preprocessing – calculating intial graftal positions
Level of Detail (LOD) – Scaling, Styling, Abstracting based on viewing distance
Rendering – Drawing the graftals and silhouette edges
The system that I intend to develop can be split in to 3 basic stages
Preprocessing
Graftals need a place to live!
The original Graftal papers used manually placed graftals
I'd like to extend that by generating the initial positions using surface- constrained particle systems
Level of Detail, also known as LOD
When the viewer gets farther away, the object should become more abstract, else we risk having an extremely noisy image
There are many ways to handle this
The straightforward approach is to reduce the number of graftals on the object, but other effects are possible
LOD can also be affected by variables like viewing angle or Time of Day
Rendering
Drawing the actual graftals and their silhouette edges
Previous systems haven't taken advantage of new hardware advancements such as programmable Geometry Shaders
Their is much room for improvement here

10. Progress Pure research for now Code framework running
Development begins next week
Eurographics 2011
Intend to submit a Short Paper
Follow progress at
realtimemike.wordpress.com
Currently in the research stage
I have a rendering framework (basically a graphics only game engine) up
I'm beginning coding the system next week
The Eurographics conference has put out a call for Short Papers
Due in December
I'll be submitting a short paper on this work (and praying it gets in!)
If you're interested at all you can follow my work at this website.
Thanks!

11. Thank You!