Multi-Class Blue Noise Sampling

1
Multi-Class Blue Noise Sampling

• Li-Yi Wei
• Microsoft Research

2
smart? ? research?

• Q not so smart?

3
Blue noise

• random uniform
• sample spacing r

4
r 0.02
single class
multi class
5
Multi-class blue noise
class 0
class 1
total set
6
Object placement
7
Sensor layout
cone/rod cells
RGB sensors
8
Color stippling
RGBCMYB dots
9
A smart? ? popular?

• simpler algorithms
• less intimidating

10
50 seconds old version
11
smart? ? research?

• Q not so smart?

12
Blue noise

• random uniform
• sample spacing r

13
r 0.02
single class
multi class
14
Multi-class blue noise
class 0
class 1
total set
15
Object placement
16
Sensor layout
cone/rod cells
RGB sensors
17
Color stippling
RGBCMYB dots
18
A smart? ? popular?

• simpler algorithms
• less intimidating

19
Old Versioncontinue joke from FF 2008
20
Blue noise distribution

• random uniform
• dart throwing
• Lloyd relaxation

21

• X not novel fast-forward SIG 2008
• X not very useful

22
overlay
X not uniform
23
Multi-class blue noise
class 0
class 1
total set
24
Object placement
25
Sensor layout
cone/rod cells
RGB sensors
26
Color stippling
RGBCMYB dots
27
Backup
28
(No Transcript)
29
Sensor layout
retina
continuous domain
discrete domain
30
(No Transcript)
31
(No Transcript)
32
Old slide from SIGGRAPH 2000
texture synthesis
input (small)
output (large)
33
New SIGGRAPH 2008 paper(just by flipping the old
slide)
inverse texture synthesis
output (small)
input (large)
34
Object placement

• Uniform distribution
• Red flower
• Yellow flower
• Entire set

35
Uniform per class
O
O
X
class 0 Poisson disk
class 1 Poisson disk
total set
36
Uniform total set
X
X
O
total set Poisson disk
class 0
class 1
37
Our method
O
O
O
class 0 Poisson disk
class 1 Poisson disk
total set Poisson disk