Large-scale, Real-world facial recognition in movie trailers

17
Large-scale, Real- world facial recognition in movie trailers Alan Wright Presentation 7

description

Large-scale, Real-world facial recognition in movie trailers. Alan Wright Presentation 7. recap of last week. Cast selector. 3. Cast selector. Retrieves cast list from Rotten Tomatoes using their API. Ignore tracks we don’t want. Type custom names. - PowerPoint PPT Presentation

Transcript of Large-scale, Real-world facial recognition in movie trailers

Page 1: Large-scale, Real-world facial recognition in movie trailers

Large-scale, Real-world facial recognition in movie

trailersAlan Wright

Presentation 7

Page 2: Large-scale, Real-world facial recognition in movie trailers

recap of last week

Page 3: Large-scale, Real-world facial recognition in movie trailers

Cast selector3

Page 4: Large-scale, Real-world facial recognition in movie trailers

Cast selector

• Retrieves cast list from Rotten Tomatoes using their API.

• Ignore tracks we don’t want.

• Type custom names.

• Allows two people to simultaneously label tracks and no labeling will be repeated.

4

Page 5: Large-scale, Real-world facial recognition in movie trailers

Cast selector

• All 2400+ tracks have now been labeled with the correct faces.

• Faces not in PubFig were still labeled.

• Easily label more tracks if new trailers are added.

• If faces are added to PubFig, the labeling will not need to be redone.

5

Page 6: Large-scale, Real-world facial recognition in movie trailers

Labeling results

• 635 Unknown tracks

• 712 PubFig tracks

• 1113 labeled tracks (faces not in PubFig)

• 4 ignored tracks.

Page 7: Large-scale, Real-world facial recognition in movie trailers

Labeling results

PubFig Ids

# of

labels

Page 8: Large-scale, Real-world facial recognition in movie trailers

Labeling results

• Katherine Heigl was labeled the most with 51 tracks.

• Each PubFig face (in the trailers) has an average of 12 tracks.

Page 9: Large-scale, Real-world facial recognition in movie trailers

Labeling Results

• The most labeled face, not in PubFig, was Edward Norton with 53 tracks.

• 218 faces were labeled, but not in PubFig.

• Average of 5 tracks per face.

Page 10: Large-scale, Real-world facial recognition in movie trailers

New pr curveAccurate with labeled faces

Page 11: Large-scale, Real-world facial recognition in movie trailers

How can we add more faces?

• Look at the distribution of faces that aren’t in PubFig

• Pick a threshold that will give us faces that appear often, and extend PubFig.

• Note: We want a good threshold because the average number of tracks per person (not in PubFig) is 5.

Page 12: Large-scale, Real-world facial recognition in movie trailers

Track distributionFaces not in PubFig

# of

labels

Face IDs

Page 13: Large-scale, Real-world facial recognition in movie trailers

Track distributionFaces not in PubFig

# of

labels

Face IDs

Threshold of 20

Page 14: Large-scale, Real-world facial recognition in movie trailers

New faces

• Choosing a threshold of 20 or more tracks gives us 9 new people:

1. Edward Norton - 53

2. Amanda Seyfried - 37

3. Jason Bateman - 34

4. Hilary Swank - 31

5. Paul Rudd - 30

6. Robert De Niro - 27 Leelee Sobiesk - 26 Dwayne Johnson - 24 Johnny Depp - 24

Page 15: Large-scale, Real-world facial recognition in movie trailers

new faces

• Downloaded images for these 9 people and added them to PubFig. (eye aligned, extracted features, etc)

Page 16: Large-scale, Real-world facial recognition in movie trailers

new labeling distribution

• 635 Unknown tracks

• 998 Extended PubFig tracks

• 827 labeled tracks (faces not in PubFig)

• 4 ignored tracks.

Page 17: Large-scale, Real-world facial recognition in movie trailers

What’s next?• Run over new supplemented data (Server will be up this afternoon)

• Implement other voting methods:

1.Logarithmic pooling

2.Borda Count

• Look at other ways to create a single confidence score for non-avg SRC and SVM methods

• Experiment with different parameters: crop, pca dimensions, features, voting