+ All Categories
Home > Documents > Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill...

Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill...

Date post: 21-Jan-2016
Category:
Upload: kelley-horton
View: 213 times
Download: 0 times
Share this document with a friend
Popular Tags:
26
Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed from Kevin Murph
Transcript
Page 1: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Context-based vision system for place and object recognition

Antonio TorralbaKevin MurphyBill FreemanMark Rubin

Presented by David LeeSome slides borrowed from Kevin Murphy

Page 2: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Object out of context

Page 3: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Object in context

Page 4: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Wearable test-bed

Page 5: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

System diagram

Page 6: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Computing the features

Page 7: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

24 filteredImages

Downs

ampl

e

to 4

x4

4x4x24=384 dim 80 dim

Page 8: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Visualizing the filter bank output

Images

80-dimensional representation

Page 9: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Place recognition system

Page 10: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Hidden Markov Model

Hidden states = location (63 values) Observations = vG

t ∈ R80

Transition model encodes topology of environment

Observation model is a mixture of Gaussians (100 views per place)

Page 11: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Hidden Markov Model

Observation Likelihood

Prediction Prior

Transition Matrix

Mixture of Gaussians MLE (counting)

Page 12: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Scene Categorization

17 Categories (Office, Corridor, Street, etc)

Train a separate HMM on category labels

Page 13: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Place recognition demo

Page 14: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Specific location

Location category

Indoor/outdoor

Ground truthSystem estimate

Performance on known env.

Page 15: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Performance on new env.

Page 16: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Comparison of features

Recognition Categorization

Page 17: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Effect of HMM on recognition

With Without(But with temporal smoothing)

Page 18: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

From place to object recognition

Page 19: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Object priming Predict object properties based on

context (top-down signals): Visual gist, vt

G

Specific Location, Qt

Kind of location, Ct

Page 20: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Object Priming

Again…MLE

Probability of object i

Probability of object i in image vi given entire video

sequence

Probability of object i Given current

observation & place

Estimate of current place

(Output of HMM)

Mixture of Gaussians

Observation Likelihood

Prior probability of object i being

in place q

Page 21: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Predicting object presence

Page 22: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

ROC curves for object detection

Page 23: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Predicting object position and scale

Page 24: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Predicting object position and scaleEstimate of

mask

Probability of an object i being present and location being q(Output of previous system)

Estimate of mask given current gist, place, and object

delta Gaussian

Page 25: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Predicted segmentation

Page 26: Context-based vision system for place and object recognition Antonio Torralba Kevin Murphy Bill Freeman Mark Rubin Presented by David Lee Some slides borrowed.

Conclusion

Real world problem (and it works!)

Uses only global feature (context)

How much did {HMM / place prior} affect{place recognition / object detection}?Can we really say “context” did the job?


Recommended