Upload
phamkhue
View
216
Download
0
Embed Size (px)
Citation preview
Introduce the field
“…a real convergence…” – “…computer vision finally becoming useful
Pikachu © Nintendo - http://www.freelargeimages.com/wp-content/uploads/2014/12/Pikachu_04.pngHoloLens © Microsoft - http://www.glassappsource.com/wp-content/uploads/2016/01/hololens.jpg
1
https://cdn1.vox-cdn.com/uploads/chorus_asset/file/6625539/pokemon-go-nick_statt-screenshots-1.0.jpg
Niantic – Ingress
Pokemon Go © Nintendo, NianticImage © Independent - https://static.independent.co.uk/s3fs-public/thumbnails/image/2016/07/08/15/pokemon-go-nick-statt-screenshots-1.0.jpghttp://www.independent.co.uk/arts-entertainment/pokemon-go-uk-release-date-and-how-to-get-it-now-on-iphone-and-android-apk-a7127326.html
Logo © Nintendo - http://assets.pokemon.com/assets/cms2/img/video-games/video-games/pokemon_go/169.jpg
2
Snapchat
Photo - http://thetechjournal.com/how-to/tutorial-face-swap-snapchat.xhtmlhttp://www.pleated-jeans.com/2016/03/02/21-snapchat-face-swaps-that-went-horribly-wrong/
3
Images © Microsofthttps://compass-ssl.surface.com/assets/13/b5/13b57736-9875-4332-af9f-dca0accbc331.jpg?n=HoloLens_Homepage_Audiences1_1920_1001.jpghttps://i.ytimg.com/vi/SKpKlh1-en0/maxresdefault.jpghttps://cdn0.vox-cdn.com/thumbor/aotdquRv2ZmMp1l-NTU_HrHcQtA=/cdn0.vox-cdn.com/uploads/chorus_asset/file/3327268/a67d3d33-e1e5-4cf7-bf3d-dbe1befc8d8c.0.jpg
Like magic! Where did that come from?!Even a company building a competitor – called MagicLeap – that has put out a few screen recordings.
4
5
6
Top: Bregler et al.Bottom – Beowolf production shots
7
Furukawa et al.
8
[Bokeloh et el. EUROGRAPHICS 2009]
9
[Ma et al. EGSR 2007]
10
Explain the diagram.
So what happens when the estimation of these models is not useful – when they are incorrect or imprecise or insufficient?
11
Question audience – what is wrong here?
Video - Playing Pokémon GO in New York - Tech Insiderhttps://www.youtube.com/watch?v=aQfkME59s9s
12
Image - http://www.pleated-jeans.com/2016/03/02/21-snapchat-face-swaps-that-went-horribly-wrong/
What went wrong here?
13
Image - http://www.pleated-jeans.com/2016/03/02/21-snapchat-face-swaps-that-went-horribly-wrong/
What went wrong here?
14
Image - http://blogs.solidworks.com/solidworksblog/wp-content/uploads/sites/2/6a00d83451706569e2017ee8115a91970d.jpg
Question audience – what is wrong here?
Next slide: but when you get it right, it can be transformative.
15
Magic Leap video - https://www.youtube.com/watch?v=kw0-JRa9n94
Where else are these models useful?
Half face shot is next.
16
https://www.wired.com/2013/03/luxion-keyshot/
Beyond augmented reality to VFX
Can anyone guess what is going on in this image? What artifact do you see?
17
Ex Machina – VFX Showreel – Double Negative -https://www.youtube.com/watch?v=qN5cjt2O6js
“A young programmer is selected to participate in a ground-breaking experiment in synthetic intelligence by evaluating the human qualities of a breath-taking humanoid A.I.”
Academy Award for best visual effects, 2016.
How did they fill in the bits _behind_ the newly transparent parts?
18
19
Image © Jason Corso - http://web.eecs.umich.edu/~jjcorso/r/career/index.html
20
21
OK, so before I give you a little introduction about myself, who do we have in the room?
Undergrads?Masters?PhDs?
And what’s your experience in visual computing?
Who has taken a graphics course?Vision course?Machine learning? Interaction?
22
I’ve been working in visual computing research now for 10 years, and in that time I’ve combined state-of-the-art computer vision techniques to create new interactivegraphics applications.
23
Well, he’s one example result.
The user provides start and end pair of images, and the system generates an
exploration of the video collection that takes in those views. This is computed
completely automatically, just from the visual information.
In this result, the system smoothly and automatically moves from looking at Big Ben to looking at the London Eye – in this composite motion made from 4 videos, the camera moves from one side of the river to the other and rotates 180 degrees to take in both views.
Tompkin et al. SIGGRAPH 2012 Videoscapes
24
Here’s what it looks like in motion.When we move through a video’s timeline, the video foci moves spatially in the context. We can see the comings and goings of the people in the scene, across different times – potentially across many months.In this way, by embedding the videos within a shared context, we can move away from the typical linear sequential presentation of videos, to build new interactions to compare and contrast spatio-temporal events.
FOCUS + CONTEXT -> great for YURT resolution demo
Tompkin et al. UIST 2013 Vidicontexts
25
Tompkin et al. UIST 2013 Vidicontexts
26
How might we paint in free space within a volume, with a hand-held pen, and have those paintings appear exactly where you placed them?
Tompkin et al. UIST 2015 5D light field pen sensing
27
Now, in my time as a masters and doctoral student, many things were hard for me.
28
Especially tough right at the beginning of your studies.
29
30
Especially tough right at the beginning of your studies.
31
32
One person is assigned at random at the beginning of the seminar to lead the discussion.Everybody leads discussion at least once in the course.The discussion leader receives a digest of the submitted questions just before the seminar.The discussion leader raises questions appropriately throughout the discussion, covers future work aspects, and finally provides a summary of the strengths and weaknesses of the techniques and of the discipline.
33
34
35
36
37
38
You spend a lot of time in high school and later on in college, under certain disciplines, learning to write.But learning to read efficiently when we move to reading academic texts is an important skill for your dissertations or thesis, and it’s not really taught as much.
39
40
PatchMatch+Global Patch Collider or TreeCANN
Image © Barnes et al., PatchMatch, SIGGRAPH 2009http://gfx.cs.princeton.edu/pubs/Barnes_2009_PAR/patchmatch.pdf
41
Stereo diagram © MATLAB - https://www.mathworks.com/matlabcentral/mlc-downloads/downloads/submissions/41739/versions/1/screenshot.jpgModel column architecture – Goesele et al. – Multi-view Stereo Revisited -http://www.cs.washington.edu/node/2322http://www.gcc.tu-darmstadt.de/media/gcc/papers/Goesele-2006-MSR.pdf
42
https://stimulant.com/depth-sensor-shootout-2/
Learning to be a depth camerahttps://www.youtube.com/watch?v=n89EsXM69d4http://research.microsoft.com/en-us/projects/depth4free/default.aspx
Project TangoVenture Beat - http://1u88jj3r4db2x4txp44yqfj1.wpengine.netdna-cdn.com/wp-content/uploads/2015/08/Project-Tango-Google2-930x585.jpg
43
Dou et al., SIGGRAPH 2016 -http://perceptiveio.com/publications/2016/7/13/fusion4d-real-time-performance-capture-of-challenging-scenes
44
Left – Shotton et al., Real-Time Human Pose Recognition in Parts from a Single Depth Image - http://research.microsoft.com/apps/pubs/default.aspx?id=145347
Right – Gall et al., http://files.is.tue.mpg.de/jgall/projects/skelsurf/skelsurf.htmlMotion Capture Using Joint Skeleton Tracking and Surface Estimation
45
Debevec et al. – Rendering Synthetic Objects into Real Scenes (SIGGRAPH 1998)https://www.fxguide.com/wp-content/uploads/2012/05/relit_cglighting.jpg?8f1314
NOTE: This is _not_ a difficult illumination estimation paper, but it is one. Uses HDR images of environment maps to light the virtual world.We will read papers on more advanced problems.
46
Image © Song et al. - http://ait.inf.ethz.ch/projects/2015/3Dfrom2D/ - Joint Estimation of 3D Hand Position and Gestures from Monocular Video for Mobile Interaction
47
Video – WIRED - https://www.youtube.com/watch?v=BLkFWq_ipCc
Has anyone heard of Magic Leap?
$800 million seed round
WIRED
48
Image © Wikipedia - https://en.wikipedia.org/wiki/Mixed_reality
49
A great literature review
50
Video – Märki et al. - https://graphics.ethz.ch/~perazzif/bvs/index.html
51
Video – Märki et al. - https://graphics.ethz.ch/~perazzif/bvs/index.html
52
Images © MIT intrinsic image database http://www.cs.toronto.edu/~rgrosse/intrinsic/Roger Grosse, Micah K. Johnson, Edward H. Adelson, and William T. Freeman, Ground truth dataset and baseline evaluations for intrinsic image algorithms, in Proceedings of the International Conference on Computer Vision (ICCV), 2009.
53
Video © Laffont et al., Coherent Intrinsic Images from Photo Collections, http://www-sop.inria.fr/reves/Basilic/2012/LBPDD12/
54
Video © Face2Face: Real-time Face Capture and Reenactment of RGB Videos -http://people.mpi-inf.mpg.de/~mzollhoef/Papers/CVPR2016_FF/page.htmlJustus Thies, Michael Zollhöfer, Marc Stamminger, Christian Theobalt, Matthias NießnerCVPR 2016
55
56
57
58
59
60
61