Distributed Ray Tracer

1
Distributed Ray Tracer

• Watkinson and Woodring

2
What is distributed ray tracing?

• Like raytracing, calculate light bounces with
  vector calculations to get shadows, reflections,
  refractions, etc.
• Sample over time and space to reduce aliasing and
  reconstruct the function the image is supposed to
  represent
• Binary operations, like nearest neighbor
  textures, produce aliasing, so can things like
  shadow in non-distributed ray tracing is it in
  the shadow or not in the shadow?

3
Steps

• Supersample for each pixel
• Sample over time
• Sample over pixel
• Sample over lens
• Raytrace
• Sample over light
• Sample over reflections
• Sample over refractions

4
Whats in our raytracer?

• No distributed raytracing or
• Antialiasing
• Soft Shadows (not best method)
• Gloss Reflections
• Blurry Translucency
• Depth of Field
• Motion Blur (toggle-able)

5
Antialiasing

• Jitter over the pixel
• Sample the colors and take the average color
• A function produces the jitter given a step size
  and side of a square

6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
Soft Shadows

• Importance cone sampling over the area light
• Find the plane the normal makes, pick a point in
  the plane, setup coordinate axes, and find a
  normally distributed point in the square from the
  center
• Pi/32 half cone angle for light sampling
• Not exactly best, should be sampling over the
  area the light takes up- pick a point in the area
  and see if it can see the object

10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
Gloss Reflections

• Importance cone sampling from reflected rays
• Same methodology as light sampling for soft
  shadows
• User input cone half angles for sampling

14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
Blurry Translucency

• Same exact method as glossy reflections except
  with refracted rays instead of reflected rays

18
(No Transcript)
19
(No Transcript)
20
(No Transcript)
21
Depth of Field

• Sample over the lens, just like sampling over the
  pixel
• Square lens, so recycle the jitter function
• The lens is located at the origin before
  transformation and the focal point is located
  somewhere down the view vector

22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
Motion Blur

• Toggle-able motion blur, because we have to
  interpolate motion at every step and retransform
  objects, can get expensive
• Jitter the frame time then interpolate position
  of the object

26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
Miscellany

• Image plane divided up into 32 segments and
  distributed across 32 processors
• Introduced faster intersection tests for spheres
  and polygons
• Had to read in two frames, only linear
  interpolation in motion
• Implemented area lights to support soft shadows