Fcv learn learned-miller

26
Computer Science Department Learning on the Fly: Rapid Adaptation to the Image Erik Learned-Miller with Vidit Jain, Gary Huang, Laura Sevilla Lara, Manju Narayana, Ben Mears

Transcript of Fcv learn learned-miller

Page 1: Fcv learn learned-miller

Computer Science Department

Learning on the Fly:Rapid Adaptation to the Image

Erik Learned-Millerwith Vidit Jain, Gary Huang,

Laura Sevilla Lara, Manju Narayana, Ben Mears

Page 2: Fcv learn learned-miller

2Learning on the Fly

“Traditional” machine learning

Learning happens from large data sets

• With labels: supervised learning

• Without labels: unsupervised learning

• Mixed labels: semi-supervised learning,transfer learning,learning from one (labeled) example,self-taught learning,domain adaptation

Page 3: Fcv learn learned-miller

3Learning on the Fly

Learning on the Fly

Given:

• A learning machine trained with traditional methods

• a single test image (no labels)

Learn from the test image!

Page 4: Fcv learn learned-miller

4Learning on the Fly

Learning on the Fly

Given:

• A learning machine trained with traditional methods

• a single test image (no labels)

Learn from the test image!

• Domain adaptation where the “domain” is the new image

• No covariate shift assumption.

• No new labels

Page 5: Fcv learn learned-miller

5Learning on the Fly

An Example in Computer Vision

Parsing Images of Architectural ScenesBerg, Grabler, and Malik ICCV 2007.

• Detect easy or “canonical” stuff.

• Use easily detected stuff to bootstrap models of harder stuff.

Page 6: Fcv learn learned-miller

6Learning on the Fly

Claim

This is so easy and routine for humans that it’s hard to realize we’re doing it.

• Another example…

Page 7: Fcv learn learned-miller

7Learning on the Fly

Learning on the fly…

Page 8: Fcv learn learned-miller

8Learning on the Fly

Learning on the fly…

Page 9: Fcv learn learned-miller

9Learning on the Fly

Learning on the fly…

Page 10: Fcv learn learned-miller

10Learning on the Fly

What about traditional methods…

Hidden Markov Model for text recognition:

• Appearance model for characters

• Language model for labels

• Use Viterbi to do joint inference

Page 11: Fcv learn learned-miller

11Learning on the Fly

What about traditional methods…

Hidden Markov Model for text recognition:

• Appearance model for characters

• Language model for labels

• Use Viterbi to do joint inference

DOESN’T WORK!

Prob( |Label=A) cannot be well estimated, fouling up the whole process.

Page 12: Fcv learn learned-miller

12Learning on the Fly

Lessons

We must assess when our models are broken, and use other methods to proceed….

• Current methods of inference assume probabilities are correct!

• “In vision, probabilities are often junk.”

• Related to similarity becoming meaningless beyond a certain distance.

Page 13: Fcv learn learned-miller

13Learning on the Fly

2 Examples

Face detection (CVPR 2011)

OCR (CVPR 2010)

Page 14: Fcv learn learned-miller

14Learning on the Fly

Preview of results: Finding false negatives

Viola-Jones Learning on the Fly

Page 15: Fcv learn learned-miller

15Learning on the Fly

Eliminating false positives

Viola-Jones Learning on the Fly

Page 16: Fcv learn learned-miller

16Learning on the Fly

Eliminating false positives

Viola-Jones Learning on the Fly

Page 17: Fcv learn learned-miller

17Learning on the Fly

Run a pre-existing detector...

Page 18: Fcv learn learned-miller

18Learning on the Fly

Run a pre-existing detector...

Key

Face

Non-face

Close to

boundary

Page 19: Fcv learn learned-miller

19Learning on the Fly

Gaussian Process Regression

negative positive

learn smooth mapping

from appearance to score

apply mapping to borderline

patches

Page 20: Fcv learn learned-miller

20Learning on the Fly

Major Performance Gains

Page 21: Fcv learn learned-miller

21Learning on the Fly

Comments

No need to retrain original detector

• It wouldn’t change anyway!

No need to access original training data

Still runs in real-time

GP regression is done for every new image.

Page 22: Fcv learn learned-miller

22Learning on the Fly

Noisy Document

We fine herefore t

linearly rolatcd to the

when this is calculated

equilibriurn. In short,

on the null-hypothesis:

Initial Transcription

Page 23: Fcv learn learned-miller

23Learning on the Fly

Premise

We would like to fine confident wordsto build a document-specific model,but it is difficult to estimate Prob(error).

However, we can bound Prob(error).

Now, select words with

• Prob(error)<epsilon.

Page 24: Fcv learn learned-miller

24Learning on the Fly

“Clean Sets”

Page 25: Fcv learn learned-miller

25Learning on the Fly

Document specific OCR

Extract clean sets (error bounded sets)

Build document-specific models from clean set characters

Reclassify other characters in document

• 30% error reduction on 56 documents.

Page 26: Fcv learn learned-miller

26Learning on the Fly

Summary

Many applications of learning on the fly.

Adaptation and bootstrapping new models is more common in human learning than is generally believed.

Starting to answer the question: “How can we do domain adaptation from a single image?”