Title: The Layout Consistent Random Field for detecting and segmenting occluded objects
1The Layout Consistent Random Fieldfor detecting
and segmenting occluded objects
John Winn Jamie Shotton
CVPR, June 2006
2LayoutCRF contributions
- Detection and segmentation
- Handles occlusion and deformation
- Multiple objects simultaneously
- Multiple classes
3Roadmap
- Related work
- Layout consistency
- Layout Consistent Random Field
- Results
4Related work constellation models
X
Fergus et al. CVPR 2003
Leibe et al. ECCV 2004
Crandall et al. ECCV 2006
Kumar et al. CVPR 2005
5Related work constellation models
X
X
X
X
Fergus et al. CVPR 2003
Leibe et al. ECCV 2004
Crandall et al. ECCV 2006
Kumar et al. CVPR 2005
6Related work windowed detectors
Classifier
Car
Localised features
Sliding window
Viola and Jones CVPR 2001 Shotton et al.
ICCV 2005
7Related work windowed detectors
Classifier
Car? Wall?
Localised features
Sliding window
Viola and Jones CVPR 2001 Shotton et al.
ICCV 2005
8Related work multiclass segmentation
TextonBoost Shotton et al. ECCV 2006
building
tree
car
road
Doesnt exploit layout of parts cant identify
object instances
Tu et al. CVPR 2003 He et al. CVPR 2004
9Roadmap
- Related work
- Layout consistency
- Layout Consistent Random Field
- Results
10Dense part labelling
- Automatic per-pixel labelling based on a grid of
parts
11Dense part labelling
Background label
12Patch-based part detector
13Decision trees
Extremely efficient at both training and test
time. e.g. takes 2ms to apply to 160x120 image
using difference of pixel intensities.
Improved performance with multiple decision trees
(random forest).
Performs as well as boosting with shared
features, but can process much more data in the
same time.
14Patch-based part detector
15Layout consistency
16Layout consistency
17Layout consistency
18Layout consistency
Neighboring pixels
(p,q)
(p,q)
(p1,q)
(p1,q-1)
Allows for deformation/rotation
(p1,q1)
19Layout consistency
Neighboring pixels
(p,q)
?
(p,q1)
(p,q)
(p1,q1)
(p-1,q1)
Layoutconsistent
20Occlusions
Not layout consistent occlusion (or invalid)
21Effect of layout consistency
Input image
Part detector output
With layout consistency
Layout consistent regions
22Roadmap
- Related work
- Layout consistency
- Layout Consistent Random Field
- Results
23Layout Consistent Random Field
Part labels h
24Layout Consistent Random Field
Part labels h
25Layout Consistent Random Field
Parameters ? ßbg , ßoe , ßco , ßiif , e0 , ?
(set by hand)
26Inference of MAP labelling
Graph cuts with customised alpha-expansion move
Part labels h
Boykov and Jolly, ICCV 2001
27Inference of MAP labelling
Graph cuts with customised alpha-expansion move
Proposed labelling
Part labels h
Boykov and Jolly, ICCV 2001
28Inference of MAP labelling
Graph cuts with customised alpha-expansion move
Expansion move not accepted
Proposed labelling
Part labels h
Boykov and Jolly, ICCV 2001
29Inference of MAP labelling
Graph cuts with customised alpha-expansion move
Proposed labelling
Part labels h
Boykov and Jolly, ICCV 2001
30Example inference
31Decision tree re-learning
- Part-labels are inferred (constrained by known
mask) and decision forest re-trained
32Limitation of layout consistency
- Allows arbitrary stretching/scaling
33Global layout
Part labels h
Global layout constraint is (weak) star-shaped
constellation model Constrains part
locationsrelative to centroid
Instance T1
Allows competition between different object
instances
34Example with global consistency
Input image
Layout consistent regions
Instance labelling
35Roadmap
- Related work
- Layout consistency
- Layout Consistent Random Field
- Results
36UIUC car database
Segmentation accuracy 96.5 pixels correct
(assessed on 20 randomly selected, hand-labelled
images)
37UIUC car database
Segmentation accuracy 96.5 pixels correct
(assessed on 20 randomly selected, hand-labelled
images)
38UIUC car database detection
Results refer to detection of unoccluded cars
only.
39Detecting heavily occluded faces
- Caltech face database with artificial occlusions
- AR face database with real occlusions
40Stability of part labelling
Part color key
41Multi-class detection
- Can extend to multiple classes with different
numbers of part labels for each class - Example building has multiple parts, other
classes have one
42Summary future directions
- Summary
- LayoutCRF achieves multi-class detection and
segmentation of occluded, deformable objects
- Future directions
- Extend to multiple viewpoints and multiple scales
- Share parts between classes
- Incorporate object context (car above road)
- Incorporate geometric cues
43Thank you
jwinn_at_microsoft.com http//johnwinn.org/
44(No Transcript)