SLIDE 1
Megastereo:
Constructing High-Resolution Stereo Panoramas
Christian Richardt 1,2 Yael Pritch 1 Henning Zimmer 1,3 Alex Sorkine-Hornung 1
1 2 3
1
2
Megastereo: Constructing High-Resolution Stereo Panoramas 1 - - PowerPoint PPT Presentation
Megastereo: Constructing High-Resolution Stereo Panoramas 1 - - PowerPoint PPT Presentation
Megastereo: Constructing High-Resolution Stereo Panoramas 1 Christian Richardt 1,2 Yael Pritch 1 2 Henning Zimmer 1,3 3 Alex Sorkine-Hornung 1 1 2 3 Structure of my talk 1. capturing stereoscopic panoramas 2. our image alignment pipeline
SLIDE 2
SLIDE 3
3
SLIDE 4
Structure of my talk
- 1. capturing stereoscopic panoramas
- 2. our image alignment pipeline
- 3. our flow-based ray interpolation
- 4. results + live demo
5
SLIDE 5
How to capture stereo panoramas?
6
SLIDE 6
How to capture stereo panoramas?
6
SLIDE 7
How to capture stereo panoramas?
10
SLIDE 8
How to capture stereo panoramas?
10
SLIDE 9
How to capture stereo panoramas?
10
SLIDE 10
11
Omnistereo panoramas
[Peleg et al., 2001]
SLIDE 11
Omnistereo panoramas
[Peleg et al., 2001]
12
SLIDE 12
Omnistereo panoramas
[Peleg et al., 2001]
12
SLIDE 13
Omnistereo panoramas
[Peleg et al., 2001]
13
SLIDE 14
Omnistereo panoramas
[Peleg et al., PAMI 2001] (our result)
14
SLIDE 15
Related work
panoramas:
– Szeliski (2006) – Brown & Lowe (2007)
generalised cameras:
– Gupta & Hartley (1997) – Zomet et al. (2003) – Yu & McMillan (2004)
multi-perspective:
– Agarwala et al. (2006) – Rav-Acha et al. (2008) – Yu et al. (2010) – Kopf et al. (2010)
stereo panoramas:
– Hum & He (1999) – Peleg et al. (2001)
image alignment:
– Lucas & Kanade (1981) – Snavely et al. (2006) – Wu et al. (2011)
image stitching:
– Burt & Adelson (1983) – Shum & Szeliski (2000) – Kang et al. (2004) – Levin et al. (2004)
15
SLIDE 16
Motivation
unresolved practical issues in stereo panoramas: visible seams misalignment vertical parallax this is unpleasant in 2D, but intolerable in stereo
16
SLIDE 17
Contributions
a general and efficient solution for creating high-quality, high-resolution (stereo) panoramas revised image stabilisation and alignment: correcting camera orientations removing undesired vertical parallax interpolating continuous ray space from discrete views: resolving sampling artefacts virtually & on the fly
17
SLIDE 18
Structure of my talk
- 1. capturing stereoscopic panoramas
- 2. our image alignment pipeline
- 3. our flow-based ray interpolation
- 4. results + live demo
18
SLIDE 19
circular motion challenging: hand-held 720 × 1280 (Canon S95)
Input video
19
SLIDE 20
Image alignment: comparison
image-based alignment
- ur alignment approach
20
SLIDE 21
Image alignment: comparison
image-based alignment
- ur alignment approach
20
SLIDE 22
Raw input video
input video
- mnistero panorama (crop)
21
SLIDE 23
Lens undistortion
undistorted images
22
- mnistero panorama (crop)
SLIDE 24
Orientation stabilisation
stabilised images
23
- mnistero panorama (crop)
SLIDE 25
Vertical parallax cancellation
compensated images
24
- mnistero panorama (crop)
SLIDE 26
Structure of my talk
- 1. capturing stereoscopic panoramas
- 2. our image alignment pipeline
- 3. our flow-based ray interpolation
- 4. results + live demo
25
SLIDE 27
Without strip blending
‘refaim’ dataset
[Rav-Acha et al., 2008]
far: duplication near: truncation
26
SLIDE 28
Linear strip blending
‘refaim’ dataset
[Rav-Acha et al., 2008]
far: duplication near: truncation
27
SLIDE 29
Duplication + truncation
near objects far objects panoramic imaging surface
28
SLIDE 30
Our flow-based ray interpolation
near objects far objects panoramic imaging surface
29
SLIDE 31
Our flow-based ray interpolation
near objects far objects panoramic imaging surface
29
SLIDE 32
Without strip blending
‘refaim’ dataset
[Rav-Acha et al., 2008]
far: duplication near: truncation
30
SLIDE 33
Linear strip blending
‘refaim’ dataset
[Rav-Acha et al., 2008]
far: duplication near: truncation
31
SLIDE 34
Our flow-based blending
‘refaim’ dataset
[Rav-Acha et al., 2008]
far: stretching near: squeezing
32
SLIDE 35
Blending comparison
33
no blending
SLIDE 36
Blending comparison
34
linear blending
SLIDE 37
Blending comparison
35
- ur flow-based blending
SLIDE 38
Structure of my talk
- 1. capturing stereoscopic panoramas
- 2. our image alignment pipeline
- 3. our flow-based ray interpolation
- 4. results + live demo
36
SLIDE 39
360 º zoom
37
SLIDE 40
360 º zoom
38
SLIDE 41
Street panorama (linear motion)
‘refaim’ dataset
[Rav-Acha et al., 2008]
39
SLIDE 42
360 º 100% zoom
140 MP stereo panorama
40
SLIDE 43
Live demo
41
SLIDE 44
Conclusion
42
a general and efficient solution for creating high-quality, high-resolution stereo panoramas Future work: extension to more general multi-perspective images handling changing exposures stereo panorama videos
richardt.name/megastereo disneyresearch.com/project/megastereo/
SLIDE 45
Conclusion
42
a general and efficient solution for creating high-quality, high-resolution stereo panoramas Future work: extension to more general multi-perspective images handling changing exposures stereo panorama videos