my group’s current research on image understanding

98
My Group’s Current Research on Image Understanding

Upload: suzan-tyler

Post on 20-Jan-2016

213 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: My Group’s Current Research on Image Understanding

My Group’s Current Research on Image Understanding

Page 2: My Group’s Current Research on Image Understanding

An image-understanding task

Page 3: My Group’s Current Research on Image Understanding

Low-level vision

Page 4: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Low-level vision

Page 5: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Simple SegmentationLow-level vision

Page 6: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Simple SegmentationLow-level vision

Object recognition

Page 7: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Simple SegmentationLow-level vision

Object recognition

High-level perception

Page 8: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Simple SegmentationLow-level vision

Object recognition

High-level perception

Pattern recognition

Page 9: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Simple SegmentationLow-level vision

Object recognition

High-level perception

Pattern recognition

Analogy-making

Page 10: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Simple SegmentationLow-level vision

Object recognition

High-level perception

Pattern recognition

“Meaning”

Analogy-making

Page 11: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Simple SegmentationLow-level vision

Object recognition

High-level perception

??? Pattern recognition

“Meaning”

Analogy-making

Page 12: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Simple SegmentationLow-level vision

Object recognition

High-level perception

Pattern recognition

“Meaning”

Analogy-making

The “SEMANTIC

GAP’

Page 13: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Simple SegmentationLow-level vision

Object recognition

High-level perception

Pattern recognition

“Meaning”

Analogy-making

HMAX model of visual cortexRiesenhuber, Poggio, et al.

The “SEMANTIC

GAP’

Page 14: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Simple SegmentationLow-level vision

Object recognition

High-level perception

Pattern recognition

“Meaning”

Analogy-making

Active Symbol Architecturefor high-level perceptionHofstadter et al.

HMAX model of visual cortexRiesenhuber, Poggio, et al.

The “SEMANTIC

GAP’

Page 15: My Group’s Current Research on Image Understanding

Color, Shape, Texture

Simple SegmentationLow-level vision

Object recognition

High-level perception

Pattern recognition

“Meaning”

Analogy-making

Active Symbol Architecturefor high-level perceptionHofstadter et al.

HMAX model of visual cortexRiesenhuber, Poggio, et al.

The “SEMANTIC

GAP’

Page 16: My Group’s Current Research on Image Understanding

The HMAX model for object recognition(Riesenhuber, Poggio, Serre, et al.)

Page 17: My Group’s Current Research on Image Understanding

1. Densely tile the image withwindows of different sizes.

2. HMAX features are computed in each window.

3. The features in eachwindow are given as inputto the trained support vector machine.

4. If the SVM returns a score above a learned threshold, then the object is said to be “detected” .

Recognition Phase

Streetscenes “scene understanding” system(Bileschi, 2006)

Page 18: My Group’s Current Research on Image Understanding

Object detection (here, “car”) with HMAX model (Bileschi, 2006)

Page 19: My Group’s Current Research on Image Understanding

Some limitations of the Streetscenes approach to scene understanding

Page 20: My Group’s Current Research on Image Understanding

Some limitations of the Streetscenes approach to scene understanding

• Requires exhaustive search for object identification and localization

Page 21: My Group’s Current Research on Image Understanding

Some limitations of the Streetscenes approach to scene understanding

• Requires exhaustive search for object identification and localization

Exhaustive search over:

Page 22: My Group’s Current Research on Image Understanding

Some limitations of the Streetscenes approach to scene understanding

• Requires exhaustive search for object identification and localization

Exhaustive search over:

• Window size and location in the image

Page 23: My Group’s Current Research on Image Understanding

Some limitations of the Streetscenes approach to scene understanding

• Requires exhaustive search for object identification and localization

Exhaustive search over:

• Window size and location in the image

• Object categories (e.g., car, pedestrian, tree, etc.)

Page 24: My Group’s Current Research on Image Understanding

Some limitations of the Streetscenes approach to scene understanding

• Requires exhaustive search for object identification and localization

Exhaustive search over:

• Window size and location in the image

• Object categories (e.g., car, pedestrian, tree, etc.)

Exhaustive use of HMAX features in each window

Page 25: My Group’s Current Research on Image Understanding

• Does not recognize spatial and abstract relationships among objects for whole scene understanding

Page 26: My Group’s Current Research on Image Understanding

• Does not recognize spatial and abstract relationships among objects for whole scene understanding

• Has no prior knowledge about object categories and their place in “conceptual space”

Page 27: My Group’s Current Research on Image Understanding

• Does not recognize spatial and abstract relationships among objects for whole scene understanding

• Has no prior knowledge about object categories and their place in “conceptual space”

• HMAX model is completely feed-forward; no feedback to allow context to aid in scene understanding.

Page 28: My Group’s Current Research on Image Understanding

Goal of our project

• Perform whole-scene interpretation without exhaustive search.

– Incorporate conceptual knowledge

– Allow feedforward and feedback modes to interact

Page 29: My Group’s Current Research on Image Understanding

Person Dog

leash attached to

walking

actionaction

holds

A Simple Semantic Network (or “Ontology”)

“Dog walking”

Page 30: My Group’s Current Research on Image Understanding

But...

http://www.dogasaur.com/blog/wp-content/uploads/2011/04/dogwalker.jpg

Page 31: My Group’s Current Research on Image Understanding

But...

http://www.vet.k-state.edu/depts/development/lifelines/images/dog_jog_1435.jpg

Page 32: My Group’s Current Research on Image Understanding

Person Dog

leash attached to

walking

actionaction

holds Dog Group

running

“Dog walking”

Page 33: My Group’s Current Research on Image Understanding

Person Dog

leash attached to

walking

actionaction

holds

running

Allowing “conceptual slippage”

“Dog walking”

Dog Group

Page 34: My Group’s Current Research on Image Understanding

But...

http://3.bp.blogspot.com/_1YuoCTv4oKQ/S71jUDm7kOI/AAAAAAAAAak/jz4Pg7zzzQ8/s1600/23743577.JPG

Page 35: My Group’s Current Research on Image Understanding

http://lh3.ggpht.com/-ZZrYWeBFTjo/SFQH_0ijwaI/AAAAAAAABjA/8nwryW2BmEw/IMG_0356.JPG

Page 36: My Group’s Current Research on Image Understanding

Person

leash attached to

walking

actionaction

holds

“Dog walking”

running

Cat

Iguana

Dog

Dog Group

Tail

Page 37: My Group’s Current Research on Image Understanding

But...

http://www.mileanhour.com/post/Dog-walking-bike.aspx

Page 38: My Group’s Current Research on Image Understanding

http://cl.jroo.me/z3/Z/e/C/d/a.aaa-Thus-walking-dog.png

Page 39: My Group’s Current Research on Image Understanding

ttp://thedaemon.com/images/DARPA_Segue_Dog.jpg

Page 40: My Group’s Current Research on Image Understanding

http://www.bikeforest.com/product45422.jpg

Page 41: My Group’s Current Research on Image Understanding

http://www.k9ring.com/blog/image.axd?picture=2010%2F3%2Fwalking_dog_from_car.jpg

Page 42: My Group’s Current Research on Image Understanding

http://www.guy-sports.com/fun_pictures/dog_walking_helicopter.jpg

Page 43: My Group’s Current Research on Image Understanding

http://static.themetapicture.com/media/funny-dog-walking-horse-leash.jpg

Page 44: My Group’s Current Research on Image Understanding

http://macwetblog.files.wordpress.com/2012/05/dog-walking.jpg

Page 45: My Group’s Current Research on Image Understanding

Person Dog

leash attached to

walking

actionaction

holds

running

Cat

Iguana

Biking

Car

Helicopter

“Dog walking”

Dog Group

Driving

Segue-ingTreadmill-ing Horse

Tail

Page 46: My Group’s Current Research on Image Understanding

Active Symbol Architecture(Hofstadter et al., 1995)

Page 47: My Group’s Current Research on Image Understanding

Active Symbol Architecture(Hofstadter et al., 1995)

• Basis for – Copycat (analogy-making), Hofstadter & Mitchell

– Tabletop (anlaogy-making), Hofstadter & French

– Metacat (analogy-making and self-awareness),

Hofstadter & Marshall

and many others…

Page 48: My Group’s Current Research on Image Understanding
Page 49: My Group’s Current Research on Image Understanding

Semantic network

Temperature

Workspace

Active Symbol Architecture(Hofstadter et al., 1995)

Perceptual agents (codelets)are “active symbols”

Page 50: My Group’s Current Research on Image Understanding

Petacat:

(Descendant of Copycat, part of the PetaVision project)

Integration of Active Symbol Architecture and HMAX

Initial task:

Decide if image is an instance of “taking a dog for a walk”, and if so, how good an instance it is.

Page 51: My Group’s Current Research on Image Understanding

Workspace

Page 52: My Group’s Current Research on Image Understanding

Semantic network

Workspace

Page 53: My Group’s Current Research on Image Understanding

taking a dog for a walk

outdoors

has location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

horse

swims

ropebelt

leashsidewalk

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

is onSpatial

Relation

Semantic Network

cat

Page 54: My Group’s Current Research on Image Understanding

Property links

Slip links

taking a dog for a walk

outdoors

has location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

horse

swims

ropebelt

leashsidewalk

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

is onSpatial

Relation

Semantic Network

cat

Page 55: My Group’s Current Research on Image Understanding

Semantic Network

taking a dog for a walk

outdoors

has location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

horse

swims

ropebelt

leashsidewalk

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

is onSpatial

Relation

cat

Page 56: My Group’s Current Research on Image Understanding

taking a dog for a walk

outdoors

has location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

horse

swims

ropebelt

leashsidewalk

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

is onSpatial

Relation

cat

Page 57: My Group’s Current Research on Image Understanding

taking a dog for a walk

outdoors

has location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

horse

swims

ropebelt

leashsidewalk

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

is onSpatial

Relation

cat

Page 58: My Group’s Current Research on Image Understanding

taking a dog for a walk

outdoors

has location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

horse

swims

ropebelt

leashsidewalk

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

is onSpatial

Relation

cat

Page 59: My Group’s Current Research on Image Understanding

taking a dog for a walkhas location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

cathorse

swims

ropebelt

leash

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

sidewalk

outdoors

is on

Spatial Relation

Page 60: My Group’s Current Research on Image Understanding

taking a dog for a walkhas location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

horse

swims

ropebelt

leash

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

is on

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

sidewalk

outdoors

Spatial Relation

cat

Page 61: My Group’s Current Research on Image Understanding

taking a dog for a walkhas location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

cathorse

swims

ropebelt

leash

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

is on

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

sidewalk

outdoors

Spatial Relation

Page 62: My Group’s Current Research on Image Understanding

taking a dog for a walkhas location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

cathorse

swims

ropebelt

leash

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

is on

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

sidewalk

outdoors

Spatial Relation

Page 63: My Group’s Current Research on Image Understanding

• Measures how well organized the program’s “understanding” is as processing proceeds

– Little organization high temperature

– Lots of organization low temperature

• Temperature feeds back to affect perceptual agents:

– High temperature low confidence in decisions decisions are made more randomly

– Low temperature high confidence in decisions decisions are made more deterministically

Temperature

Page 64: My Group’s Current Research on Image Understanding

Input image

Page 65: My Group’s Current Research on Image Understanding

Input image Weak segmentation

Page 66: My Group’s Current Research on Image Understanding

Input image Weak segmentation

Location “heat map”(probability distribution over pixel locations)_

++++

+

Page 67: My Group’s Current Research on Image Understanding

Input image Weak segmentation

Location “heat map”(probability distribution over pixel locations)_

++++

+

Scale “heat map”(probability distribution over scales at each pixel location)

Page 68: My Group’s Current Research on Image Understanding

Dog?

Scout codelets: Send C1 features in window to corresponding SVM.If positive result, post builder codelet with urgency equal to SVM’sconfidence.

Page 69: My Group’s Current Research on Image Understanding

Dog? Dog?

Person?

Scout codelets: Send C1 features in window to corresponding SVM.If positive result, post builder codelet with urgency equal to SVM’sconfidence.

Page 70: My Group’s Current Research on Image Understanding

Dog? Dog?

Sidewalk?

Person?

Scout codelets: Send C1 features in window to corresponding SVM.If positive result, post builder codelet with urgency equal to SVM’sconfidence.

Page 71: My Group’s Current Research on Image Understanding

Dog? Dog?

Sidewalk?

Person?

Dog?

Outdoors?

Scout codelets: Send C1 features in window to corresponding SVM.If positive result, post builder codelet with urgency equal to SVM’sconfidence.

Page 72: My Group’s Current Research on Image Understanding

Dog?negative Dog?

negative

Sidewalk?positive: 0.4

Person?negative

Outdoors?positive: 0.7

Scout codelets: Send C1 features in window to corresponding SVM.If positive result, post builder codelet with urgency equal to SVM’sconfidence.

Dog?positive: 0.8

Page 73: My Group’s Current Research on Image Understanding

Builder codelets: Ask HMAX to compute C2 features using prototype shapesspecific to the object class, and send them to corresponding SVM. If positive, decide to build structure with probability equal to SVM confidence. Break competing structures if necessary.

Dog?negative Dog?

negative

Sidewalk?positive: 0.4

Person?negative

Outdoors?positive: 0.7

Dog?positive: 0.8

Page 74: My Group’s Current Research on Image Understanding

Outdoors

Dog

Page 75: My Group’s Current Research on Image Understanding

taking a dog for a walkhas location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

cathorse

swims

ropebelt

leash

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

is on

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

sidewalk

outdoors

Spatial Relation

Page 76: My Group’s Current Research on Image Understanding

Object-specific heat maps are updated.

+

Dog

Person heat map

+

Page 77: My Group’s Current Research on Image Understanding

Object-specific heat maps are updated.

+

Dog

Person heat map

+

Dog

Person?Person?

Page 78: My Group’s Current Research on Image Understanding

Object-specific heat maps are updated.

As codelets build structure, heat maps

are continually updated to reflect prior

(learned) expectations about location

and scale as a function of location and

scale of “built” objects.

+

Dog

+

Person heat map

Person?Person?

Page 79: My Group’s Current Research on Image Understanding

Dog? Dog

Leash?

OutdoorsLeash?

Sidewalk?

Person?

Person?

Page 80: My Group’s Current Research on Image Understanding

Dog

PersonStrength: 0.75

Outdoors

Sidewalk

PersonStrength: 0.6

Page 81: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Sidewalk

Page 82: My Group’s Current Research on Image Understanding

taking a dog for a walkhas location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

cathorse

swims

ropebelt

leash

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

is on

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

sidewalk

outdoors

Spatial Relation

Page 83: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Sidewalk

Leash?

Leash?

Dog?

Sidewalk?

Dog?

Rope?

Page 84: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Sidewalk

Leash

Dog(weak)

Page 85: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Sidewalk

Leash

Dog(weak)

Dog(strong)

Page 86: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Sidewalk

Leash

Dog

Page 87: My Group’s Current Research on Image Understanding

taking a dog for a walkhas location

persondog

has action

is on

is touching

has component

aroad

abeach

trail

drives

runsflies

cathorse

swims

ropebelt

leash

string

walkswalks

is in front of

has location

has action

has component

has componenthas component

stands

is on

sits

is in front of

is touching

is behind

is next to

is on

agrass

is touching

Object

Action

indoors

sidewalk

outdoors

Spatial Relation

Page 88: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Sidewalk

Leash

Dog

Once objects begin to be built, relation and grouping codelets can run on them.

is next to

is next to

Dog group

Page 89: My Group’s Current Research on Image Understanding

Once objects begin to be built, relation and grouping codelets can run on them.

Dog

PersonOutdoors

Sidewalk

Dog

is next to

is next to

Dog group

Leash

Page 90: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Sidewalk

Dog

is next to

is next to

Dog group

is next to

Leash

Once objects begin to be built, relation and grouping codelets can run on them.

Page 91: My Group’s Current Research on Image Understanding

How Petacat makes a final decision

Temperature

taking a dog for a walk

Dog

PersonOutdoorsLeash

Dog

is next to

is next to

Dog group Sidewalk

is next to

Page 92: My Group’s Current Research on Image Understanding

How Petacat makes a final decision

Temperature

taking a dog for a walk

Dog

PersonOutdoorsLeash

Dog

is next to

is next to

Dog group Sidewalk

“Situation” codelet is more likely to run when temperature is low.

is next to

Page 93: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Leash

Dog

is next to

is next to

Dog group

is next to

Sidewalk

Situation codelet tries to match prototypical situation with existing workspace structures, possibly allowing slippages.

Page 94: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Leash

Dog

is next to

is next to

Dog group

Sidewalk

person

taking a dog for a walk

leash

dog

outdoors

is next to

has componenthas component

has component

has location

is in front of

Situation codelet tries to match prototypical situation with existing workspace structures, possibly allowing slippages.

Page 95: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Leash

Dog

is next to

is next to

Dog group

person

taking a dog for a walk

leash

dog

outdoors

is next to

has componenthas component

has component

has location

is in front of

is next toDog group

Sidewalk

Page 96: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Leash

Dog

is next to

is next to

Dog group

person

taking a dog for a walk

leash

dog

outdoors

is next to

has componenthas component

has component

has location

is in front of

is next toDog group

If resulting temperature is low enough, classify scene as positive

Sidewalk

Page 97: My Group’s Current Research on Image Understanding

Dog

PersonOutdoors

Leash

Dog

is next to

is next to

Dog group Sidewalk

person

taking a dog for a walk

leash

dog

outdoors

is next to

has componenthas component

has component

has location

is in front of

is next toDog group

If situation codelet fails enough times or does not run for a long time,program has increasing chance of ending with negative classification.

If resulting temperature is low enough, classify scene as positive

Page 98: My Group’s Current Research on Image Understanding

Temperature at the end of the run gives a measure of how good an instance the picture is (e.g., of the “dog walking” situation).

Temperature