Rendering Competition 2007/2008
Animation
Created
by:
Torsten Gründel
Name of
the animation:
The Venus Garden
Animation
Link:
Used Techniques:
Description:
The first part takes place in front of
the garden, where the camera moves through some fog on a sidewalk with
lamps on both sides. It ends in front of a gate with two dragon
statues, where the gate opens and gives a first sight into the garden
and the silhouette of a Venus statue. For the dragon statues I used a
procedural texture to give them a marble like look (more artificial
marble). The lamp model in this part of the animation is taken from http://blender-archi.tuxfamily.org/Main_Page
, the dragon statues are a high resolution model, taken from http://graphics.stanford.edu/data/3Dscanrep/.
In the second part, the camera moves
through the gate and ends up in a misty garden, where some light from
above shines through the clouds. The light is a textured light, which
simulates the clouds on the sky, that let the light only in some areas
pass through. The intensity and the direction of the light source moves
and let some halo effects appear that illuminate the Venus statue that
is in the center of the second part of the animation. I also used a
small aperture here, in order to get a small Depth of Field effect
which let the light strikes appear more unsharpened. The Venus model
was taken from
http://graphics.im.ntu.edu.tw/~robin/courses/cg03/model/
This is a sequence of pictures taken from the animation:
SAH-KDTree
The SAH-KDTree follows the approach of the KDTree that was given in the
MicroTrace solution, by successively splitting up the 3D space by
splitting planes. The main difference lies in the position of these
splitting planes. It is not calculated as the median, but is determined
by the surface area heuristic. It approximates the traversal costs of a
node created by a split plane by the surface area of the resulting node
and the number of triangles in the potential child. I used the
algorithm given by
Ingo Wald
and
Vlastimil Havran, which
allows it to build up the tree in O(n log(n)) time.
This table shows the differences between the creation and traversal of
the Median-KDTree and my SAH-KDTree. The pictures that were rendered
contained only the given model and had a resolution of 800 x 600. I
supersampled the scenes with 16 rays per pixel. The models were taken
from
http://graphics.stanford.edu/data/3Dscanrep/
|
Creation of
SAH-KDTree
|
Traversal of
SAH-KDTree
|
Creation of
Median-KDTree
|
Traversal of
Median-KDTree
|
bunny.obj (69,451 Triangles)
|
19.918 sec
|
32.293 sec
|
1.366 sec
|
2 min 12.588
sec
|
dragon.obj (202,520 Triangles)
|
1 min 31.937
sec
|
1 min 23.584
sec
|
4.289 sec
|
11 min 4.056
sec
|
My implementation uses a new class, which provides the methods to
generate a set of possible split events for a set of Triangles. For
these events, the split plane is calculated, using the surface area
heuristic, that provides the best result according to the heuristic.
Then the Triangles are split up in the two childs. This is done
recursively until the costs to split up the set of triangles is more
costly than to build a leaf node, just stores the triangles. All these
calculations follow the pseudo code given in the paper of
Ingo Wald and
Vlastimil Havran :
"œOn building fast kd-Trees for Ray
Tracing, and on doing that in O(N log N)".
Source Code:
Bump Mapping
Bump mapping is a technique that allows objects to look like they have
more geometric details than they really have. This is reached by
modifying the normals of a surfaces according to a height map texture.
This effect is frequently used in my scene. All surfaces expect the
dragon and the Venus statues have a height map assigned, so that they
look more realistic.
These two pictures show the wall in my scene with and without a height
map assigned to the objects. On the left side, the same texture that is
used to texture the object is used as height map. On the right side, no
height texture was used.
For the implementation I wrote a new Shader, which calculates for the
hit point of the ray with the object the partial derivatives at that
point. These partial derivatives are computed using the values of the
height map at the point. After that, the normal is perturbed according
to these derivative values in the corresponding directions.
The Source Code:
Procedural Shading
Procedural Shading is the process of not using a texture that is not
given by a picture, but the color at a point is determined by a
mathematical function. In my case, I used 3D Perlin noise in order to
generate a shader that produces marble like textures on objects.
Therefore the noise values were used to determine the color for a
point. The noise determined two clorors, which were weighted also by
the noise value.
The effect can be seen on the dragon
statues in the first part of the scene. Also the Venus statue uses such
a shader. The following picture shows the right dragon in the scene on
its own. Although the used colors are not common in nature, they
give a good expression of the abilities of the procedural shader.
My implementation uses a noise generator class, which generates 3D
Perlin Noise. The code is oriented by the book of
Matt Pharr : "Physically Based Rendering".
Furthermore I wrote a shader that uses the Noise value to determine the
used colors at a pixel (out of chose which are used for the marble) and
the weight of each color.
The Source Code:
Depth of Field
Depth of Field is the effect that only a particular region of a picture
is in focus, as it is the case in real life cameras. Using this effect,
only object in a plane, defined by a focus length are sharp. All other
objects, that lie before or behind that plane of focus are blurred.
Since for this effect not a pinhole model is use, but a camera model in
which there is a lens with some aperture, more than one ray is
responsible for the color of a pixel, where all rays, "going out from
the pixel through the lens" meet each other on one point on the plane
of focus. The size of the aperture can have dramatic effects on the
image. The following images use different apertures (from left to
right: 0.01, 0.05, 0.18).
|
|
In order to get good images at large aperture sizes, on has to send a
lot of rays per pixel into the scene. The following two images use both
a aperture of 0.05, but in the left image 9 rays are send out per
pixel, whereas in the right image 100 rays per pixel are send into the
scene.
In my scene I used a very small aperture in order to the the halo
effect in the second part of the animation not to be too sharp and let
the light beams be a bit more blurred.
For the implementation I wrote a new Camera class, which has an
aperture-size and a focal length. The focal length is used in order to
determine the plane of focus (which is in my case in camera
z-direction). The aperture-size is used in order to distribute all rays
that should be used to color a pixel evenly on a concentric disk with
radius of the aperture-size.
The Source Code:
Volume Scattering
Volume scattering is the effect of simulating the traversal of light
rays through volumes and simulate the effects that particles in this
volume would have on the ray and thus on the scene. So effects like
smoke or fog can be simulated using this technique. Since there are a
numerous of effects that could be applied and the numerical effort for
a complex volume scattering simulation can be extremely high, I focused
on two effects: Emission and Scattering
- Emission is the effect of energy that is added to the environment
from luminous particles
- Scattering is the effect of how light heading in one direction is
scatterid to other directions due to collisions with particles
These three pictures show the first part of my scene, without any
volume, with a volume that only applies emission effects and finally a
picture that contains an additional volume that also considers
scattering. One can see, how the lights in the lamps only are really
recognised
when they are in the volume, and how the dragons can't be seen very
clearly if the light rays have to travel through the volume to reach
them.
For the implementation I used two kinds of objects: VolumeIntegrators
and VolumeRegions. VolumeIntegrators are added to the scene and rays
that are traced are intersect with the Volumes as if they were Objects.
The Integrators are responsible to apply emission, scattering and other
effects. The VolumeRegions are given to the VolumeIntegrators. They
provide the values for the calculations of the Integrators and the
intersection tests are forwarded by the Integrators to its regions. I
implemented two classes of integrators, one which only considers
emission effects (this one is used in the picture in the middle) and an
integrator that considers a single scattering effect. Therefore the
volume is sampled and at each sample point a light ray is send to a
light source to determine the scattering influence at the sample point.
This effect is used in the right picture. This single scattering allows
it to generate volumetric shaddows, as can be seen in the following
picture.
|
I also implemented two kinds of regions, one that is homogeneous and
thus simulates the same amount of particles and an density region,
which provides a method the give points different particle density
values.
The effect of volume scattering can be seen everywhere is my animation.
Since the fog in the first part of the animation, as well as the halo
effect of the second part of the animation are created by using Volumes.
The Source Code:
Textured Light Source
In order to simulate clouds on the sky that let some light pass
through, I implemented a textured light source. Although the classname
let one imagine something else (I just forgot to rename it after tried
to make a textured quad area light source), this is a point light
source, that has a direction and a texture assigned to it. The concept
follows the one of the projective camera model. The direction is points
to the center of the texture "image" and all light/shadow rays that are
illuminated by this light source are attenuated by the texture values.
In my case the texture is the 2D perlin noise function that produces
cloud like shaders (see the CloudShader in MicroTrace).
In my animation I used this light source to simulate some cloud like
shape, that attenuates the light beams from above. By modifying a
coefficient for the attenuation, I changed the amount of light passing
through the attenuation texture of the light source. The left picture
shows the illumination at the beginning and the right picture the
illumination at the end of the animation and the manipluation of the
attenuation coefficient. (Please note, that I also shifted the
direction if the light source during the animation. This can also be
seen in the picture below).
The Source Code:
