CVPR 2006 Highlights

1
CVPR 2006 Highlights
New York University, June 17-22

• Vaibhav Vaish

2
Conference Statistics

• 318 papers (28 acceptance)
• 54 oral presentations (4.7)
• 1136 submissions
• 30 area chairs, 560 reviewers
• 1200 attendees (30 increase)
• Free dinner on last day

3
Awards

• Honored 5 champion reviewers
• Best Paper
• Putting Objects in Perspective
• D. Hoiem, A. Efros, M. Herbert
• Honorable mention Incremental Learning of Object
  Detectors Using a Visual Shape Alphabet
• A. Opelt, A. Pinz, A. Zisserman
• Best Poster TBA.

4
Longuet-Higgins Prize (CVPR 96)

• Neural Network-Based Face Detection
• H. Rowley, S. Baluja, T. Kanade
• Combining Greyvalue Invariants with Local
  Constraints for Object Recognition
• C. Schmid, R. Mohr

5
Workshop Highlights

• 25 Years of RANSAC
• Keynote Robert Bolles (co-inventor of RANSAC)
• 2 Keynotes by Shree Nayar (PROCAMS, Medical
  Imaging workshop)
• Projector defocus
• Separating direct and indirect illumination

Do NOT miss this at SIGGRAPH!
6
Scheduling
Orals I 90 min
Posters I 210 min
Posters 2 210 min
Time
Orals 2 90 min

• Oral presentations recorded, broadcast live
• To be put online (somewhere, sometime)

7
Papers I Liked

• Papers from Stanford
• Fun with digital photos and video
• Computational imaging and sensors
• Why Bill Gates is rich
• Obituary 3D Reconstruction
• Visual words for recognition

8
Papers from Stanford

• A Dynamic Bayesian Network Model for Autonomous
  3D Reconstruction from a Single Indoor Image
• E. Delage, H. Lee, Andrew Ng
• Learning Object Shape From Drawings to Images
• G. Elidan, Geremy Heitz, Daphne Koller
• Object Pose Detection in Range Scan Data
• Jim Rodgers, Dragomir Anguelov, H Pang, Daphne
  Koller
• A Comparison and Evaluation of Multi-View Stereo
  Algorithms
• S. Seitz, B. Curless, J. Diebel, D. Scharstein,
  R. Szeliski
• Reconstructing Occluded Surfaces blah

9
Papers from Stanford

• A Dynamic Bayesian Network Model for Autonomous
  3D Reconstruction from a Single Indoor Image
• E. Delage, H. Lee, Andrew Ng
• Learning Object Shape From Drawings to Images
• G. Elidan, Geremy Heitz, Daphne Koller
• Object Pose Detection in Range Scan Data
• Jim Rodgers, Dragomir Anguelov, H Pang, Daphne
  Koller
• A Comparison and Evaluation of Multi-View Stereo
  Algorithms
• S. Seitz, B. Curless, J. Diebel, D. Scharstein,
  R. Szeliski
• Reconstructing Occluded Surfaces blah

10
Papers I Liked

• Papers from Stanford
• Fun with digital photos and video
• Computational imaging and sensors
• Obituary 3D Reconstruction
• Visual words for recognition

11
Making a Long Video ShortDynamic Video Synopsis

• A. Rav-Acha, Yael Pritch, Shmuel Peleg.

• Video Summary
• Short
• Informative
• Accurate
• Seamless

12
Making a Long Video ShortDynamic Video Synopsis

• A. Rav-Acha, Yael Pritch, Shmuel Peleg.

Input Video
Summary Video
More demos
13
Making a Long Video ShortDynamic Video Synopsis

• Find regions of activity
• Compute summary using MRF optimization

14
What Makes A High Quality Photo ?

• The Design of High-Level Features for Photo
  Quality Assessment
• Yan Ke, Xiaoou Tang, Feng Jing

15
Some Ranking Results

• Error rate (snapshot vs professional) 24

16
What Makes A High Quality Photo ?

• Pros vs Point-and-shooters
• Simplicity
• (Sur)realism
• Basic Technique
• Features (a subset)
• Lack of blur
• Spatial edge distribution
• Color, brightness, contrast, hue count
• Learn from http//DPChallenge.com

17
Picture Collage

• J Wang, J Sun, L Quan, Xiaoou Tang, H Shum

18
Picture Collage

• Maximize informative regions, minimize blank
  space
• Optimize using random grid sampling (Bayesian
  framework)

19
Papers I Liked

• Papers from Stanford
• Fun with digital photos and video
• Computational imaging and sensors
• Why Bill Gates is rich
• Obituary 3D Reconstruction
• Visual words for recognition

20
Bilayer Segmentation of Live Video

• A. Criminisi, G. Cross, A. Blake, V. Kolmogorov
  Link

• Goals
• Single camera
• Real-time (no optic flow!)
• Good looking results

21
How it works

• Priors, priors, priors and priors
• Temporal continuity
• Spatial coherence
• Color likelihood
• Motion likelihood
• Learning
• Fast approximate binary graph cut

22
A Closed Form Solution to Natural Image Matting

• Anat Levin, Dani Lischinski, Yair Weiss
• Idea in a small window, colors lie on a line in
  color space
• Find alpha by minimizing aT L a
• Eigenvectors of L suggest good scribbles

23
Lensless Imaging with a Controllable Aperture

• Assaf Zomet, Shree Nayar

24
Other Papers

• Instant 3Descatter
• Tali Treibitz, Yoav Schechner
• Blind Haze Separation
• S Shwartz, E Namer, Yoav Schechner
• Space-time Video Montage
• H Kang, Y Matsuhita, Xiaoou Tang, Xue-Quan Chen

25
Papers I Liked

• Papers from Stanford
• Fun with digital photos and video
• Computational imaging and sensors
• Obituary 3D Reconstruction
• Visual words for recognition

26
Multi-View Stereo Evaluation

• S. Seitz, B. Curless, J Diebel, D Scharstein, R
  Szeliski
• http//vision.middlebury.edu/mview

27
Multi-View Stereo Taxonomy

• Scene representation
• Photo-consistency measure
• Visibility model
• Shape prior
• Reconstruction algorithm
• Initialization

28
Multi-View Stereo Evaluation

• Metrics
• Accuracy
• Completeness
• Running time
• Renderings
• Conclusions
• Most work pretty well
• Having lots of views enables simpler algorithms
  Multi-view Stereo Revisited, Goesele et al

29
Upcoming Deadlines

• December 3rd, 2006.
• CVPR 2007, Minneapolis
• March 2007
• ICCV 2007, Rio de Janeiro
• CVPR 2008 in Anchorage, Alaska

30
(No Transcript)