1. Compressive Displays
Combining Optical Fabrication, Computation, and Perceptual
Tricks to Build the Displays of the Future
Gordon Wetzstein
MIT Media Lab
media.mit.edu/~gordonw
displayblocks.org
14. Some Depth Cues of the HVS
binocular disparity
convergence
current glasses-based (stereoscopic) displays
near-term: compressive light field displays
longer-term: holographic displays
motion parallax
accommodation/blur
15.
16. Parallax Barriers – Ives 1903
barrier
2D display
• low resolution & very dim
• switchable 2D/3D with LCDs
17. lenslets
Integral Imaging – Lippmann 1908
2D display
• brighter than parallax barriers
• always low resolution, even for 2D
19. Structural Formula for Compressive Display Innovation
M
N
C
N
U
C
O
C
multiplexing
nonlinear
compression
optimization
user experience
content
community
next
22. Space vs Time
visual acuity / retina display: 20/20 = 1 arc minute
@ 10” need >300 dpi
multiplexing requires 300++ dpi
flicker fusion threshold: 20-60 Hz (depends on light)
LCDs: 120-240 Hz
DMDs, ferroelectric LCDs: KHz
much easier with currently available hardware!
29. linear
l = p1+p2
p2
mask 2
mask 1
not blocked!
p1
Additive Layers / Most Volumetric Displays
e.g., LEDs or transparent OLEDs
30. Holography – Nonlinear Interaction
plane wave
emitted wavefront: U(x)
received intensity:
I(x) = T {U(x)}
2
= Re {T {U(x)}} + Im {T {U(x)}}
2
hologram
2
screen or retina
31. Holography – Coupling
plane wave
æ
x'ö æ
x'ö
W (x, u = l sin(q )) = ò t ç x + ÷ t ç x - ÷ e2 p ix'u dx '
è
2ø è
2ø
Fourier transform of all points interacting
with each other!
hologram
nonlinear interaction & pixel coupling!
32. Layered 3D – SIGGRAPH 2011
attenuation layers with
spacers
backlight
33. Layered 3D – SIGGRAPH 2011
mask K
…
mask 2
mask 1
34. Layered 3D – SIGGRAPH 2011
l = p1*p2*…*pK
mask K
pK
…
mask 2
mask 1
p2
p1
35. Layered 3D – SIGGRAPH 2011
mask K
…
mask 2
mask 1
56. Compressive
Optics
16:9 media window
This slide has a
Display-adaptive
Compression
Computed Tomography
4D Light Field
Nonnegative Tensor
Factorization
Uniform or
Directional Backlight
Stacked Layers
(LCDs or Transparencies)
Observer = Decoder
57. Give those Pixels a Break!
Applied Mathematics
Benefits for Optics & Electronics
• sparse optimization
• fewer pixels
• low-rank factorization
• relaxation on refresh rate
• …
• thinner form factors
• …
58. From Conventional to Compressive Displays
mask 2
mask 1
Conventional Parallax Barriers
t
Parallax Barriers
1903
Time-Shifted
Parallax Barriers 2007
HR3D
SIG Asia 2010
t
Layered 3D
SIGGRAPH 2011
Tensor Displays
SIGGRAPH 2012
t
59. From Conventional to Compressive Displays
mask 2
mask 1
Time-shifted Parallax Barriers [Kim et al. 2007]
High Resolution through High Speed
t
Parallax Barriers
1903
Time-Shifted
Parallax Barriers 2007
HR3D
SIG Asia 2010
t
Layered 3D
SIGGRAPH 2011
Tensor Displays
SIGGRAPH 2012
t
60. From Conventional to Compressive Displays
Perceptual Integration
time
Time-shifted Parallax Barriers [Kim et al. 2007]
High Resolution through High Speed
t
Parallax Barriers
1903
Time-Shifted
Parallax Barriers 2007
HR3D
SIG Asia 2010
t
Layered 3D
SIGGRAPH 2011
Tensor Displays
SIGGRAPH 2012
t
61. From Conventional to Compressive Displays
mask 2
mask 1
High-Rank 3D [Lanman et al., SIGGRAPH Asia 2010]
Compression in Time – Nonnegative Matrix Factorization
t
Parallax Barriers
1903
Time-Shifted
Parallax Barriers 2007
HR3D
SIG Asia 2010
t
Layered 3D
SIGGRAPH 2011
Tensor Displays
SIGGRAPH 2012
t
62. From Conventional to Compressive Displays
Perceptual Integration
time
High-Rank 3D [Lanman et al., SIGGRAPH Asia 2010]
Compression in Time – Nonnegative Matrix Factorization
t
Parallax Barriers
1903
Time-Shifted
Parallax Barriers 2007
HR3D
SIG Asia 2010
t
Layered 3D
SIGGRAPH 2011
Tensor Displays
SIGGRAPH 2012
t
63. From Conventional to Compressive Displays
mask K
…
mask 2
mask 1
Layered 3D [Wetzstein et al., SIGGRAPH 2011]
Compression in Pixels & Depth – Computed Tomography
t
Parallax Barriers
1903
Time-Shifted
Parallax Barriers 2007
HR3D
SIG Asia 2010
t
Layered 3D
SIGGRAPH 2011
Tensor Displays
SIGGRAPH 2012
t
64. From Conventional to Compressive Displays
mask K
…
mask 2
mask 1
Layered 3D [Wetzstein et al., SIGGRAPH 2011]
Compression in Pixels & Depth – Computed Tomography
t
Parallax Barriers
1903
Time-Shifted
Parallax Barriers 2007
HR3D
SIG Asia 2010
t
Layered 3D
SIGGRAPH 2011
Tensor Displays
SIGGRAPH 2012
t
65. From Conventional to Compressive Displays
mask K
…
mask 2
mask 1
Layered 3D [Wetzstein et al., SIGGRAPH 2011]
Compression in Pixels – Computed Tomography
t
Parallax Barriers
1903
Time-Shifted
Parallax Barriers 2007
HR3D
SIG Asia 2010
t
Layered 3D
SIGGRAPH 2011
Tensor Displays
SIGGRAPH 2012
t
66. From Conventional to Compressive Displays
…
Perceptual Integrat
…
…
time
Tensor Displays
Compression in Time & Pixels –Tensor Factorization
t
Parallax Barriers
1903
Time-Shifted
Parallax Barriers 2007
HR3D
SIG Asia 2010
t
Layered 3D
SIGGRAPH 2011
Tensor Displays
SIGGRAPH 2012
t
67. From Conventional to Compressive Displays
…
Perceptual Integrat
…
…
time
Tensor Displays – Multilayer & Directional Backlighting
t
Parallax Barriers
1903
Time-Shifted
Parallax Barriers 2007
HR3D
SIG Asia 2010
t
Layered 3D
SIGGRAPH 2011
Tensor Displays
SIGGRAPH 2012
t
68. From Conventional to Compressive Displays
thin!
Perceptual Integrat
time
Tensor Displays – Directional Backlighting
t
Parallax Barriers
1903
Time-Shifted
Parallax Barriers 2007
HR3D
SIG Asia 2010
t
Layered 3D
SIGGRAPH 2011
Tensor Displays
SIGGRAPH 2012
t
75. Tomographic Light Field Synthesis
image formation
virtual planes
x
L(x, q ) = e
-
ò c m (r )dr
log ( L ( x, q )) = - ò m (r)dr
attenuation volume
c
backlight
tomographic synthesis
2D light field
argmin log( L) P
0
x
2
2
76. Tomographic Polarization Field Synthesis
virtual planes
f3
LCD 3
f2
LCD 2
image formation
K
x
f1
Q(x, q ) = åfk (x, q )
k=1
L(x, q ) = sin 2 (Q(x,q ))
LCD 1
tomographic synthesis
backlight
Q(x,q ) = ±sin-1
2D light field
(
argmin Q - Pf
fmin £f £fmax
x
)
L(x, q ) mod p
2
2
77. Low-rank Light Field Factorization
light field
two-layer light field display
L(
front layer
fm(2)(
rear layer
backlight
fm(1)(
78. Low-rank Light Field Factorization
two-layer light field display
front layer
front layer
fm(2)(
rear layer
backlight
Lanman et al. – SIGGRAPH Asia 2010
fm(1)(
rear layer
L(
`
rank-1
79. Low-rank Light Field Factorization
high-speed LCDs = perceptual average
two-layer light field display
L(
front layer
fm(2)(
rear layer
backlight
Lanman et al. – SIGGRAPH Asia 2010
fm(1)(
`
rank-4
80. Low-rank Light Field Factorization
high-speed LCDs = perceptual integration
G
~
L
FG
Lanman et al. – SIGGRAPH Asia 2010
F
~
`
L
rank-4
81. Low-rank Light Field Factorization
arg min L - FG Wfield F,G ³ 0
, for
emitted light
2
F,G
`
L
=
G
F
82. Low-rank Light Field Factorization
light field
light field tensor
multi-layer light field display
L(
fm(3)(
middle layer
fm(2)(
rear layer
backlight
fm(1)(
rear layer
front layer
L(
83. Low-rank Light Field Factorization
Target Light Field Tensor
Rank-M Approximation
nonnegative tensor
factorization (NTF)
+ ... +
+
frame 1
perceptual
integration
frame 2
frame M
84. Low-rank Light Field Factorization
nonlinear (multilinear)
optimization problem
iterative update rules
(see paper for details)
Efficient GPU Implementation
forward projection (multiview rendering)
back projection (projective texture mapping)
101. Compressive Light Field Photography – SIG 2013
Exploit
redundancy
Sparsify
Preserve
information
Compressive
Reconstruction
Light Field Atoms
Mask-based
Light Field Coding
102. Prototype Setup with a Variable Mask
LCoS
Virtual sensor
Polarizing
Imaging Beamsplitter
Lens
Camera
Image sensor
103.
104. Display-Adaptive Image Synthesis – SIG 2013
“conventional pipeline”
1. render trillions of rays
2. large-scale optimization
3. display pixel states
116. Structural Formula for Compressive Display Innovation
M
N
C
N
U
C
O
C
multiplexing
nonlinear
compression
optimization
user experience
content
community
next
117. Gordon Wetzstein
MIT Media Lab
media.mit.edu/~gordonw
displayblocks.org
collaborators:
Matt Hirsch (MIT)
Doug Lanman (NVIDIA)
Andrew Maimone (UNC)
Felix Heide (UBC)
Fu-Chung Huang (UC Berkeley)
Vincent Lee (MIT)
James Gregson (UBC)
Brian Barsky (UC Berkeley)
Henry Fuchs (UNC)
Ramesh Raskar (MIT)
Wolfgang Heidrich (UBC)
code, data & instructions online
use Layered 3D in your class!
Editor's Notes
Use the original video clips in higher res and the same layout but better quality and not this!Get rid of the gray background as well
Use the original video clips in higher res and the same layout but better quality and not this!Get rid of the gray background as well