Large Image Databases and Small Codes for Object Recognition Rob Fergus (NYU) Antonio Torralba (MIT)...

transcript

Large Image Databases and Small Codes

for Object Recognition

Rob Fergus (NYU)Antonio Torralba (MIT)Yair Weiss (Hebrew U.)

William T. Freeman (MIT)

Object Recognition

Pixels Description of scene contents

Amazing resource Maybe we can use it for recognition?

Large image datasets

Internet contains billions of images

Web image dataset• 79.3 million images• Collected using image

search engines• List of nouns taken from

Wordnet

• 32x32resolution

a-bomba-horizona._conan_doylea._e._burnsidea._e._housmana._e._kennellya.e.a_batterya_cappella_singinga_horizon

• Example first 10:

See “80 Million Tiny images” TR

Noisy Output from Image Search Engines

Nearest-Neighbor methods for recognition

# images

Issues

• Need lots of data• How to find neighbors quickly?• What distance metric to use?

• How to transfer labels from neighbors to query image?

• What happens if labels are unreliable?

Overview

1.Fast retrieval using compact codes

2.Recognition using neighbors with unreliable labels

Overview

Binary codes for images

• Want images with similar contentto have similar binary codes

• Use Hamming distance between codes– Number of bit flips– E.g.:

• Semantic Hashing [Salakhutdinov & Hinton, 2007]– Text documents

Ham_Dist(10001010,10001110)=1

Ham_Dist(10001010,11101110)=3

Binary codes for images

• Permits fast lookup via hashing– Each code is a memory address– Find neighbors by exploring Hamming

ball around query address– Lookup time depends on

radius of ball, NOT on # data points

Address Space

Query ImageSemantic HashFunction

Semantically similar images

Query address

n ‘0

Compact Binary Codes

• Google has few billion images (109)• Big PC has ~10 Gbytes (1011 bits)• Codes must fit in memory (disk too slow) Budget of 102 bits/image

• 1 Megapixel image is 107 bits• 32x32 color image is 104 bits Semantic hash function must also reduce

dimensionality

Code requirements

• Preserves neighborhood structureof input space

• Very compact (<102 bits/image)• Fast to compute

Three approaches:1.Locality Sensitive Hashing (LSH)2.Boosting3.Restricted Boltzmann Machines (RBM’s)

Image representation: Gist vectors• Pixels not a convenient representation• Use GIST descriptor instead• 512 dimensions/image• L2 distance btw. Gist vectors not bad

substitute for human perceptual distance

Oliva & Torralba, IJCV 2001

1. Locality Sensitive Hashing• Gionis, A. & Indyk, P. & Motwani, R. (1999)• Take random projections of data• Quantize each projection with few bits

•For our N bit code:– Compute first N PCA components of data– Each random projection must be linear combination of the N PCA components

2. Boosting

• Modified form of BoostSSC [Shaknarovich, Viola & Darrell, 2003]

• GentleBoost with exponential loss• Positive examples are pairs of neighbors• Negative examples are pairs of unrelated images• Each binary regression stump examines a single

coordinate in input pair, comparing to some learnt threshold to see that they agree

3. Restricted Boltzmann Machine (RBM)

Hidden units: h

Visible units: v

Symmetric weights w

Parameters: Weights w Biases b

• Network of binary stochastic units• Hinton & Salakhutdinov, Science 2006

RBM architecture

Hidden units: h

Visible units: v

Symmetric weights w

Learn weights and biases using Contrastive Divergence

Parameters: Weights w Biases b

• Network of binary stochastic units• Hinton & Salakhutdinov, Science 2006

Convenient conditional distributions:

Multi-Layer RBM: non-linear dimensionality reduction

Input Gist vector (512 dimensions)

Layer 1

Layer 2

Layer 3

Output binary code (N dimensions)

Training RBM models• Two phases

1. Pre-training – Unsupervised– Use Contrastive Divergence to learn weights and biases– Gets parameters to right ballpark

2. Fine-tuning– Can make use of label information– No longer stochastic– Back-propagate error to update parameters– Moves parameters to local minimum

Greedy pre-training (Unsupervised)

Layer 1

Activations of hidden units from layer 1 (512 dimensions)

Layer 2

Activations of hidden units from layer 2 (256 dimensions)

Layer 3

Backpropagation using Neighborhood Components Analysis objective

Layer 1

Layer 2

Layer 3

Output binary code (N dimensions)

Neighborhood Components Analysis• Goldberger, Roweis, Salakhutdinov & Hinton, NIPS 2004

Labels defined ininput space (high dim.)

Points in output space (low dimensional)

Output of RBMW are RBM weights

Pulls nearby pointsOF SAME CLASScloser

Push nearby pointsOF DIFFERENT CLASSaway

Pulls nearby pointsOF SAME CLASScloser

Preserves neighborhood structure of original, high dimensional, space

Push nearby pointsOF DIFFERENT CLASSaway

Two test datasets1. LabelMe– 22,000 images (20,000 train | 2,000 test)– Ground truth segmentations for all– Can define ground truth distance btw. images using these

segmentations – Per pixel labels [SUPERVISED]

2. Web data– 79.3 million images– Collected from Internet– No labels, so use L2 distance btw. GIST vectors as

ground truth distance [UNSUPERVISED]– Noisy image labels only

Retrieval Experiments

Examples of LabelMe retrieval• 12 closest neighbors under different distance metrics

LabelMe Retrieval

Size of retrieval set % o

0 2,000 10,000 20,0000

LabelMe Retrieval

Size of retrieval set % o

0 2,000 10,000 20,0000

Number of bits% o

Web images retrieval%

Size of retrieval set

Web images retrieval

Size of retrieval set

Examples of Web retrieval

• 12 neighbors using different distance metrics

Retrieval Timings

LabelMe Recognition examples

LabelMe Recognition

Overview

Label Assignment

• Retrieval gives set of nearby images• How to compute label?

• Issues:– Labeling noise– Keywords can be very specific

• e.g. yellowfin tuna

Query Grover Cleveland Linnet Birdcage Chiefs Casing

Neighbors

Wordnet – a Lexical Dictionary

Synonyms/Hypernyms (Ordered by Estimated Frequency) of noun aardvark

Sense 1aardvark, ant bear, anteater, Orycteropus afer => placental, placental mammal, eutherian, eutherian mammal => mammal => vertebrate, craniate => chordate => animal, animate being, beast, brute, creature

=> organism, being => living thing, animate thing => object, physical object => entity

http://wordnet.princeton.edu/

Wordnet Hierarchy

Synonyms/Hypernyms (Ordered by Estimated Frequency) of noun aardvark

Sense 1aardvark, ant bear, anteater, Orycteropus afer => placental, placental mammal, eutherian, eutherian mammal => mammal => vertebrate, craniate => chordate => animal, animate being, beast, brute, creature

=> organism, being => living thing, animate thing => object, physical object => entity

• Convert graph structure into tree by taking most common meaning

Wordnet Voting Scheme

Ground truth

One image – one vote

Classification at Multiple Semantic Levels

Votes:

Animal 6Person 33Plant 5Device 3Administrative4Others 22

Votes:

Living 44Artifact 9Land 3Region 7Others 10

Wordnet Voting Scheme

Person Recognition

• 23% of all imagesin dataset containpeople

• Wide range ofposes: not justfrontal faces

Person Recognition – Test Set

• 1016 images fromAltavista using“person” query

• High res and 32x32available

• Disjoint from 79million tiny images

Person Recognition• Task: person in image or not? • 64-bit RBM code trained on web data (Weak image

labels only)

Viola-JonesViola-Jones

Object Classification• Hand-designed metric

Distribution of objects in the world

10,000

10 100 1,000

L a b e lM e s t a t i s t i c s

object rank

10% of the classes account for 93% of the labels!

1500 classes with lessthan 100 samples

Number oflabeledsamples

LabelMe dataset

Conclusions

• Possible to build compact codes for retrieval– # bits seems to depend on dataset– Much room for improvement– Use JPEG coefficients as input to RBM

• Can do interesting things with lots of data– What would happen with Google’s ~ 2 billion

images?– Video data

Large Image Databases and Small Codes for Object Recognition Rob Fergus (NYU) Antonio Torralba (MIT)...

Documents