Docstoc

Learned_Miller

Document Sample
Learned_Miller Powered By Docstoc
					                    Faces in the Wild
                    Detection, Alignment and Recognition
                             of Real World Faces
                             Erik Learned-Miller
                  with Vidit Jain, Gary Huang, Andras Ferencz, et al.




Computer Science Department
 Is Face Recognition Solved?




Computer Science               2
 Is Face Recognition Solved?

          “100% Accuracy in Automatic Face Recognition” [!!!]
                                      Science 25 January 2008




            QuickTime™ and a
  TIF F (Uncompressed) decompressor
    are needed to see this picture.




Computer Science                                                3
 Is Face Recognition Solved?

          “100% Accuracy in Automatic Face Recognition” [!!!]
                                      Science 25 January 2008




            QuickTime™ and a
  TIF F (Uncompressed) decompressor
    are needed to see this picture.   A history of overstated results.




Computer Science                                                         4
 The Truth
   Many different face recognition problems
      • Out of context, accuracy is meaningless!
   Many problems are REALLY HARD!
      • For some problems
               state of the art is 70% or worse!
   We have a long way to go!




Computer Science                                   5
 Face Recognition at UMass
     Labeled Faces in the Wild
     The Detection-Alignment-Recognition pipeline
     Congealing and automatic face alignment
     Hyper-features for face recognition
     New directions in recognition




Computer Science                                     6
 Labeled Faces in the Wild
                   http://vis-www.cs.umass.edu/lfw/




                                        QuickTime™ and a
                               TIFF (Uncompressed) decompressor
                                  are needed to see this picture.




Computer Science                                                    7
 The Many Faces of Face Recognition




                             Quic kTime™ and a                      Quic kTime™ and a
                   TIFF (Uncompres sed) dec ompress or    TIFF (Uncompres sed) dec ompress or
                      are needed to s ee this pic ture.      are needed to s ee this pic ture.




                        Labeled Faces in the Wild




Computer Science                                                                                 8
 The Many Faces of Face Recognition




                             Quic kTime™ and a                      Quic kTime™ and a
                   TIFF (Uncompres sed) dec ompress or    TIFF (Uncompres sed) dec ompress or
                      are needed to s ee this pic ture.      are needed to s ee this pic ture.




                      Labeled Faces in the Wild




Computer Science                                                                                 9
 The Many Faces of Face Recognition




                             Quic kTime™ and a                      Quic kTime™ and a
                   TIFF (Uncompres sed) dec ompress or    TIFF (Uncompres sed) dec ompress or
                      are needed to s ee this pic ture.      are needed to s ee this pic ture.




                         Labeled Faces in the Wild




Computer Science                                                                                 10
 The Many Faces of Face Recognition




                             Quic kTime™ and a                      Quic kTime™ and a
                   TIFF (Uncompres sed) dec ompress or    TIFF (Uncompres sed) dec ompress or
                      are needed to s ee this pic ture.      are needed to s ee this pic ture.




                         Labeled Faces in the Wild




Computer Science                                                                                 11
 The Many Faces of Face Recognition




                                                                    QuickTime™ a nd a
                                                          TIFF (Uncompressed) decompressor
                             Quic kTime™ and a               are need ed to see this picture.

                   TIFF (Uncompres sed) dec ompress or
                      are needed to s ee this pic ture.




                    Labeled Faces in the Wild




Computer Science                                                                                12
 Labeled Faces in the Wild
     13,233 images, with name of each person
     5749 people
     1680 people with 2 or more images

     Designed for the “unseen pair matching problem”.
       •   Train on matched or mismatched pairs.
       •   Test on never-before-seen pairs.
     Distinct from problems with “galleries” or training data for each
      target image.
     Best accuracy: currently about 73%!




Computer Science                                                          13
 Detection-Alignment-Recognition Pipeline

  Detection
                                           Alignment                              Recognition

                                                                                                   “Same”
                                                 QuickTime™ an d a
                                             TIFF (LZW) decompressor
             QuickTime™ and a             are need ed to see this p icture .
    TIFF (Uncompressed) decompressor
       are need ed to see this picture.

                                                                                      QuickTime™ an d a
                                                                                  TIFF (LZW) decompressor
                                                                               are need ed to see this p icture .




                                                                                                                               Qu i ckTi me ™ an d a
                                                                                                                    TIFF (Un co mp re s se d) de co mp re s so r
                                                                                                                       a re ne ed ed to se e thi s p i ctu re .




Computer Science                                                                                                                                 14
 Detection-Alignment-Recognition Pipeline

  Detection
                                               Alignment                              Recognition

                                                                                                       “Same”
                                                     QuickTime™ an d a
                                                 TIFF (LZW) decompressor
             QuickTime™ and a                 are need ed to see this p icture .
    TIFF (Uncompressed) decompressor
       are need ed to see this picture.

                                                                                          QuickTime™ an d a
                                                                                      TIFF (LZW) decompressor
                                                                                   are need ed to see this p icture .




                                                                                                                                   Qu i ckTi me ™ an d a
                                                                                                                        TIFF (Un co mp re s se d) de co mp re s so r
                                                                                                                           a re ne ed ed to se e thi s p i ctu re .




                                     Parts should work together.

Computer Science                                                                                                                                     15
 Labeled Faces in the Wild
   All images are output of a standard
         face detector.
   Also provides aligned images.
   Consequence: any face recognition algorithm
         that works well on LFW can easily be turned
         into a complete system.




Computer Science                                       16
 Congealing (CVPR 2000)




                          QuickTime™ and a
                           decompressor
                   are neede d to see this picture.




Computer Science                                      17
  Criterion of Joint Alignment

   Minimize sum of pixel stack
    entropies by transforming
    each image.




                                  A pixel stack




Computer Science                             18
 Congealing Complex Images

           Window around pixel   SIFT vector and clusters



                                                            SIFT clusters




                                         vector representing
                                         probability of each cluster,
                                         or “mixture” of clusters



Computer Science                                                            19
       QuickTime™ an d a
Microsoft Video 1 decomp resso r
 are need ed to see this p icture .
Crash Course on Martian Identification
                                      Martian training set




   Test: Find Bob after one meeting
                                               =

                   ?
                                                =

        Bob

                                                 =

Computer Science                                             21
 Training Data




       “same”
                          QuickTime™ an d a
                      TIFF (LZW) decomp resso r
                   are need ed to see this picture .
                                                       “different”




Computer Science                                                     22
 General Approach to Hyper-feature method
   Carefully align objects
   Develop a patch-based model of
    image differences.
   Score match/mismatch based on patch
    differences.




Computer Science                            23
 Three Models
  1. Universal patch model:
      P(patchDistance|same)
      P(patchDistance|different)
  2. Spatially dependent patch model:
      P(patchDistance |same,x,y)
      P(patchDistance |different,x,y)
  3. Hyper-feature dependent model:
      1. P(patchDistance |same,x,y,appearance)
      2. P(patchDistance |different,x,y,appearance)




Computer Science                                      24
 Universal Patch Model
                   A single P(dist | same) for all patches




   Different blue patches are evidence against a match!

Computer Science                                             25
 Spatial Patch Model
 P(dist|same,x1,y1) estimated separately from P(dist|same,x2,y2)




        Greatly increases discriminativeness of model.

Computer Science                                                   26
 Hyper-Feature Patch Model




           Is the patch from a matching face going to
           match this patch?



Computer Science                                        27
 Hyper-Feature Patch Model




           Is the patch from a matching face going to
           match this patch? Probably yes



Computer Science                                        28
 Hyper-Feature Patch Model




                             What about
                             this patch?




Computer Science                           29
 Hyper-Feature Patch Model




                             What about
                             this patch?
                             Probably
                             not.


Computer Science                           30
   Ridiculous Errors from the World’s Best Unconstrained Face
   Recognition System




       Quic kTime™ and Quic kTime™ and a
                           a
               TIFF (Uncompres sed) dec ompress or
F (Uncompres sed) dec ompress or
are needed to s eeare needed to s ee this pic ture.
                   this pic ture.




  Computer Science                                          31
 Ridiculous Errors from the World’s Best Unconstrained Face
 Recognition System




                   Quic kTime™ and a                      Quic kTime™ and a
         TIFF (Uncompres sed) dec ompress or
            are needed to s ee this pic ture.   TIFF (Uncompres sed) dec ompress or
                                                   are needed to s ee this pic ture.




Computer Science                                                                       32
 The New Mission: Estimate Higher Level Features




                            QuickTime™ an d a
                        TIFF (LZW) decomp resso r
                     are need ed to see this picture.




Computer Science                                        33
 The New Mission: Estimate Higher Level Features




                                                        Can we guess
                            QuickTime™ an d a
                        TIFF (LZW) decomp resso r
                     are need ed to see this picture.
                                                           pose?




Computer Science                                                       34
 The New Mission: Estimate Higher Level Features




                                                        Can we guess
                            QuickTime™ an d a
                        TIFF (LZW) decomp resso r
                     are need ed to see this picture.
                                                          gender?




Computer Science                                                       35
 The New Mission: Estimate Higher Level Features




                                                     Can we guess
                            QuickTime™ an d a
                        TIFF (LZW) decomp resso r  degree of balding,
                     are need ed to see this picture.

                                                     beardedness,
                                                      moustache?




Computer Science                                                        36
 The New Mission: Estimate Higher Level Features




                                                    Can we say that
                            QuickTime™ an d a
                        TIFF (LZW) decomp resso r
                     are need ed to see this picture.
                                                     none of these
                                                     individuals are
                                                   the same person?




Computer Science                                                       37
 What can we do with a good segmentation?




                          QuickTime™ an d a
                      TIFF (LZW) decomp resso r
                   are need ed to see this picture.




Computer Science                                      38
 CRF Segmentations




                          QuickTime™ and a
                      TIFF (LZW) decompressor
                   are neede d to see this picture.




Computer Science                                      39
 CRF Segmentations




                          QuickTime™ and a
                      TIFF (LZW) decompressor
                   are neede d to see this picture.




Computer Science                                      40
 Who’s This?




Computer Science   41
 Who’s This?




                            QuickTime™ and a
                   TIFF (Uncompressed) decompressor
                     are needed to see this picture.




Computer Science                                       42
 Who’s This?




               QuickTime™ and a a
                QuickTime™ and
      TIFF (Uncompressed) decompressor
            (Uncompressed) decompressor
       TIFF needed to see this picture.
        are needed to see this picture.
         are
              from www.coolopticalillusions.com


Computer Science                                  43
                              Thanks




Computer Science Department

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:3
posted:3/6/2012
language:
pages:44