34
Large-Scale, Real- World Face Recognition in Movie Trailers Alan Wright

Large-Scale, Real-World Face Recognition in Movie Trailers

  • Upload
    soleil

  • View
    27

  • Download
    0

Embed Size (px)

DESCRIPTION

Large-Scale, Real-World Face Recognition in Movie Trailers. Alan Wright. Plan of Attack. Extract Facial Tracks from videos Extract the features from the facial tracks. Build framework to load and test data. Begin with baseline testing (Sparse, min, mean, etc) Algorithm development…. - PowerPoint PPT Presentation

Citation preview

Page 1: Large-Scale, Real-World Face Recognition in Movie Trailers

Large-Scale, Real-World Face Recognition in Movie

Trailers

Alan Wright

Page 2: Large-Scale, Real-World Face Recognition in Movie Trailers

Plan of Attack

• Extract Facial Tracks from videos• Extract the features from the facial tracks.• Build framework to load and test data.• Begin with baseline testing (Sparse, min,

mean, etc)• Algorithm development…

Page 3: Large-Scale, Real-World Face Recognition in Movie Trailers

Dataset

• 2010 Movie Trailerso 305 videoso Avg. of 625 faces per trailero Avg. of 23 face trackers per trailer

• PubFigo 58,787 photoso 200 people o Avg. of 295 photos per person.

Page 4: Large-Scale, Real-World Face Recognition in Movie Trailers

Dataset

Facetrack: Dictionary:

Page 5: Large-Scale, Real-World Face Recognition in Movie Trailers

Face Tracks

• We have:

• Instead of:

Page 6: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination Approach

+ x2 + x3

+ x4 + x5 + x6

+ x7 + x8 + x9

Test Image

= x1

Training Images

Page 7: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination Approach

y

Testing

A=

=

Training

x

Coefficients

Page 8: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination & Face Tracks

Page 9: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination & Face Tracks

Page 10: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination & Face Tracks

Page 11: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination & Face Tracks

Page 12: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination & Face Tracks

Page 13: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination & Face Tracks

Page 14: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination & Face tracks

Page 15: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination & Face Tracks

Page 16: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination & Face Tracks

Page 17: Large-Scale, Real-World Face Recognition in Movie Trailers

Linear Combination & Face Tracks

Sylvester Stallone

Page 18: Large-Scale, Real-World Face Recognition in Movie Trailers

Variations

• Blurry images within face tracks

• Unknowns

• Facial Expressions

Page 19: Large-Scale, Real-World Face Recognition in Movie Trailers

Variations

• Pose

• Occlusions

Page 20: Large-Scale, Real-World Face Recognition in Movie Trailers

Preliminary Testing

• Ran SRC facial recognition on extracted facial tracks from Date Night.

• Choose the name that has the most “votes” in the track.

• Looking strictly at Steve Carell and Tina Fey tracks…

Page 21: Large-Scale, Real-World Face Recognition in Movie Trailers

Results

• Correctly assessed 15 out of 34 face tracks correctly (44.12% accuracy).

• 6 out of 18 Tina Fey face tracks (33.33% accuracy).

• 9 out of 16 Steve Carell face tracks (56.25% accuracy).

Page 22: Large-Scale, Real-World Face Recognition in Movie Trailers

Correct Result

• Track 7- IDed CORRECTLY as: Tina Feyo Alyssa Milano had 2 voteso Drew Barrymore had 4 voteso Jennifer Love Hewitt had 1 voteso Salma Hayek had 3 voteso Tina Fey had 22 votes

• 32 frames in this track

• 68.75% identified as Tina Fey

Page 23: Large-Scale, Real-World Face Recognition in Movie Trailers

Correct Result

Page 24: Large-Scale, Real-World Face Recognition in Movie Trailers

Correct Result

• Good facial crop.

• Many frames.

• Good lighting

• Non blurry

• Generally straightforward pose

Page 25: Large-Scale, Real-World Face Recognition in Movie Trailers

Incorrect Result (1)

• Track 9- IDed as: Gael Garcia Bernalo Clive Owen had 2 voteso Cristiano Ronaldo had 1 voteso Gael Garcia Bernal had 3 voteso Steve Carell had 2 votes

Page 26: Large-Scale, Real-World Face Recognition in Movie Trailers

Incorrect Result (1)

Page 27: Large-Scale, Real-World Face Recognition in Movie Trailers

Incorrect Result (1)

• Blurry

• Not a great crop

• Only 7 frames

Page 28: Large-Scale, Real-World Face Recognition in Movie Trailers

Incorrect Result (2)

• Track 26 - IDed as: Nicole Kidmano David Duchovny had 2 voteso Lucy Liu had 1 voteso Michael Bloomberg had 1 voteso Nicole Kidman had 3 voteso Salma Hayek had 1 votes

Page 29: Large-Scale, Real-World Face Recognition in Movie Trailers

Incorrect Result (2)

Page 30: Large-Scale, Real-World Face Recognition in Movie Trailers

Incorrect Result (2)

• Profile view

• Poor crop

• Only 8 frames.

Page 31: Large-Scale, Real-World Face Recognition in Movie Trailers

Incorrect Result (3)

• track 40 - IDed as: Keira Knightleyo Adriana Lima had 6 voteso Cindy Crawford had 8 voteso Eliot Spitzer had 1 voteso Eva Mendes had 1 voteso Gillian Anderson had 1 voteso Keira Knightley had 18 voteso Meg Ryan had 3 voteso Minnie Driver had 1 voteso Salma Hayek had 3 voteso Tina Fey had 15 votes

Page 32: Large-Scale, Real-World Face Recognition in Movie Trailers

Incorrect Result (3)

Page 33: Large-Scale, Real-World Face Recognition in Movie Trailers

Incorrect Result (3)

• Dark lighting

• Pose variation

• Motion blur toward the end

• 57 frames

Page 34: Large-Scale, Real-World Face Recognition in Movie Trailers

What’s next?

• Continue to look at the coefficient vector.

• Look into confidence of each choice (SCI and residual error)

• Apply different facial recognition algorithms.

• Best way to choose a name for each track? (Mode, strongest confidence, etc.)