ICCV 2005 Overview

1
ICCV 2005 Overview

• Derek Hoiem

2
PASCAL Dataset
Motorbikes
Cars
People
Bicycles
3
LabelMe Dataset
4
LabelMe Dataset
5
Perceptual scale and its applications Wang
Bahrami Zhu
6
Perceptual scale and its applications Wang
Bahrami Zhu
7
Perceptual scale and its applications Wang
Bahrami Zhu

• Applications
• Tracking over scales
• Adaptive image display

8
Globally optimal solutions for energy
minimization in stereo using reweighted belief
propagation Meltzer Yanover Weiss

• Find global optimum for stereo pairs using Tree
  Reweighted BP
• Result graph cuts also close to global minimum
  problem is in energy function itself

9
Globally optimal solutions for energy
minimization in stereo using reweighted belief
propagation Meltzer Yanover Weiss
10
Learning layered motion segmentations of video
Kumar Torr Zisserman

• Efficient loopy BP to get good initial estimate
• Alpha expansion to get final solution
• Complicated and slow but accurate

11
Learning layered motion segmentations of video
Kumar Torr Zisserman
12
Closely coupled object detection and segmentation
Zhao Davis
13
Closely coupled object detection and segmentation
Zhao Davis
14
Closely coupled object detection and segmentation
Zhao Davis
15
Closely coupled object detection and segmentation
Zhao Davis
16
Creating efficient codebooks for visual
recognition Jurie Triggs

• Like textons but for objects
• Mean-shift based rather than k-means based
  (clustering of patches)
• Conclusion works much better than keypoints and
  better than k-means
• Thorough evaluation

17
Creating efficient codebooks for visual
recognition Jurie Triggs
18
LOCUS learning object classes with unsupervised
segmentation Winn Jojic
19
LOCUS learning object classes with unsupervised
segmentation Winn Jojic
20
LOCUS learning object classes with unsupervised
segmentation Winn Jojic
21
Regular polygon detection Barnes Loy Shaw
Robles-Kelly

• Hacky method to find regular polygons in image

22
Space-time scene manifolds - Wexler Simakov

• Does alignment and mosaicing by finding
  non-linear space-time manifold in video
• Camera moves mostly in one direction
• Cut across the other direction, preserving
  appearance

23
Space-time scene manifolds - Wexler Simakov
24
Space-time scene manifolds - Wexler Simakov
25
Vector boosting for rotation invariant face
detection - Huang Ai Li Lao

• Extension of RealBoost to ouptut vector, rather
  than scalar
• Real time, amazing results

26
Vector boosting for rotation invariant face
detection - Huang Ai Li Lao
27
Convex grouping combining boundary and region
information - Stahl Wang
28
Learning hierarchical models of scenes, objects,
parts Sudderth Torralba Freeman Willsky

• Objects have shared parts which have appearances
• Scenes contain objects

29
Actions as Space-Time Shapes Blank Gorelick
Shechtman Irani Basri

• Compute some features over space-time shape and
  do NN

30
Consistent segmentation for optical flow
estimation Zitnick Jojic Kang
31
Consistent segmentation for optical flow
estimation Zitnick Jojic Kang
32
A generative/discriminative learning algorithm
for image classification Li Shapiro Bilmes

• Generative process produces fixed length
  descriptor of image (clustering)
• Discriminative process performs image retrieval

33
Bottom-up/Top-Down image parsing by attribute
graph Han Zhu
34
Bottom-up/Top-Down image parsing by attribute
graph Han Zhu
35
Learning object categories from Google's image
search Fergus Fei-Fei Perona Zisserman

• Translation-Scale Invariant pLSA