Sébastien Marcel (marcel@idiap.ch)

Sébastien Marcel - Databases of features for face verification

1. Public Features

1.1. XM2VTS Database
You can find the full XM2VTS database here.
Face localization data

manually annotated (ground truth) eye coordinates [download here (44 Kb)],

manually annotated (ground truth) facial points (markup 68) from Tim Cootes [download here (1 Mb)],

automatically localized eye coordinates using IDIAP frontal face detector [download here (46 Kb)],

automatically localized facial points (markup 68) using Tim Cootes AAM [download here (1 Mb)].

Face features
32x40 faces (32 pixels width and 40 pixels height)extracted using:

manually annotated eye coordinates [download here (16 Mb)].

Example of a 32x40 face image:

64x80 faces (64 pixels width and 80 pixels height) extracted using:

manually annotated eye coordinates [download here (67 Mb)],

manually annotated AAM facial points (download here (67 Mb)),

automatically localized using IDIAP frontal face detector (download here (66 Mb)),

automatically localized using Tim Cootes AAM (download here (67 Mb)).

Example of a 64x80 face image and its mirror:
DCTmod2 features computed from 64x80 faces extracted using:

manually annotated eye coordinates [download here (67 Mb)],

automatically localized using IDIAP frontal face detector (download here (67 Mb)),

automatically localized using Tim Cootes AAM (download here (67 Mb)).

Example of DCTmod2 features:
  > readbindata face64x80-eyecenter-dctmod2/003_1_1.bindata -verbose

  Reading bindata file (face64x80-eyecenter-dctmod2/003_1_1.bindata)
     n_inputs = 18
     n_patterns = 442

  91.1631 170.294 289.711 29.3509 -177.184 30.9695 43.6017 18.375 -43.1352 -72.5007 -34.1996 95.375 -675.25 -709.538 -236.498 -458.25 12.4165 -395.5
  -20.7292 -120.017 433.946 43.2115 9.74572 19.3821 10.9883 -8.62497 57.3052 -23.0004 24.4954 137.625 -623.125 -884.753 168.88 -60.25 99.0401 -31.7631
  -0.826204 -14.8498 373.174 34.0132 59.0986 -15.0796 -6.72669 -2.87498 -15.5066 -3.34879 -9.58199 60.875 -889.125 -762.948 116.006 -10.5 232.216 -43.0322
  ...
  30.7062 -12.1224 -4.58457 0.211894 -12.7375 -47.7896 -1.78277 -8.37497 -22.776 -11.1757 5.76193 2.12503 -206.25 -146.207 30.4936 -70.5 -10.0598 -150.858
  3.91689 23.1488 23.067 -5.99534 -21.8645 -21.3802 -1.93932 -2.99996 42.0267 45.1222 1.09895 3.50003 -98 -268.161 -153.063 130.75 24.4313 -80.9474
  -10.6045 -74.8917 -28.1095 -47.9205 55.668 161.845 8.13169 -9.37496 -94.9448 -68.0458 63.1574 -30.375 -75.75 16.734 154.607 49.75 -133.277 251.089
  
PCA features computed from 64x80 faces extracted using:

manually annotated eye coordinates [download here (4 Mb)],

automatically localized using Tim Cootes AAM (download here (4 Mb)).

Example of PCA features:
  > readbindata face64x80-eyecenter-xm2vts-lp-svd-pca243/003_1_1.bindata 
  
  Reading bindata file (face64x80-eyecenter-xm2vts-lp-svd-pca243/003_1_1.bindata)
   n_inputs = 243
   n_patterns = 2
  
  -6.38467 -0.89267 -6.83655 0.959056 2.43493 0.885989 -1.23909 1.06012 -0.044583 1.36756 -2.37928 0.13299 -1.74708 ...
  -6.38467 -0.892671 -6.83655 0.959055 2.43493 0.885991 -1.23909 -1.06012 -0.0445835 1.36756 -2.37928 0.132991 1.74708 ...
  
LDA features computed from 64x80 faces projected into the above PCA subspace extracted using:

manually annotated eye coordinates [download here (4 Mb)],

automatically localized using Tim Cootes AAM (download here (4 Mb)).

Example of LDA features:
  > readbindata face64x80-eyecenter-xm2vts-lp-svd-pca243-xm2vts-lp-qz-lda-221/003_1_1.bindata
 
  Reading bindata file (face64x80-eyecenter-xm2vts-lp-svd-pca243-xm2vts-lp-qz-lda-221/003_1_1.bindata)
   n_inputs = 221
   n_patterns = 2

  -0.535495 -0.739477 -0.521744 -1.26252 -1.89246 -1.51992 0.926259 0.579374 -0.521441 ...
  -0.535497 -0.739478 -0.521745 -1.26252 -1.89246 -1.51992 0.92626 0.579373 -0.521441 ...
  

1.2. BANCA (English) Database
You can find the full BANCA database here.
Face localization data

manually annotated (ground truth) eye coordinates [download here (126 Kb)],

automatically localized eye coordinates using IDIAP frontal face detector [download here (124 Kb)].

Face features
64x80 faces (64 pixels width and 80 pixels height) extracted using:

manually annotated eye coordinates [download here (184 Mb)],

automatically localized using IDIAP frontal face detector (download here (182 Mb)).

Example of a 64x80 face image:
DCTmod2 features computed from 64x80 faces extracted using:

manually annotated eye coordinates [download here (187 Mb)],

automatically localized using IDIAP frontal face detector (download here (184 Mb)).

Example of DCTmod2 features:
  > readbindata face64x80-eyecenter-dctmod2/1001_f_g1_s01_1001_en_1.bindata -verbose

  Reading bindata file (face64x80-eyecenter-dctmod2/1001_f_g1_s01_1001_en_1.bindata)
     n_inputs = 18
     n_patterns = 442

  -16.824 37.0966 -90.1795 -50.6635 -52.0446 12.5372 31.0372 3.25004 -3.05168 5.73744 -16.7674 40.7501 673.25 351.151 2.44213 -208.625 -80.2399 67.7665
  43.0481 -58.1493 -55.3523 -32.7057 3.60612 8.46008 27.4554 -15 32.0554 -31.3063 -20.841 3.50004 542.625 219.255 140.965 -208.25 74.4628 -41.4705
  -33.1039 6.38256 -28.9601 -9.82813 -37.5585 -3.91435 -1.26219 -4.49997 -17.862 36.0135 -8.9982 -33.9999 314.875 184.379 114.148 -243.125 118.432 63.6622
  ...
  -56.6357 -16.632 20.7506 14.2004 13.8859 16.0772 7.0101 3.75005 20.4905 13.0674 -2.04542 1.00003 -177.5 -58.3145 122.432 -175.375 -6.34681 186.848
  17.5025 42.3468 0.928702 6.49415 -4.69587 -24.2607 -3.72963 9.12502 -8.30821 -12.6506 13.6802 -4.87496 -169.375 -44.4116 -122.438 -462.375 -43.8963 125.556
  -24.1542 3.88939 -0.552799 0.801103 -0.116961 16.2115 0.690048 23.875 38.9277 4.92049 -3.78196 -4.12498 -53 -28.7525 -24.9355 -441.625 -38.3524 -140.069
  

2. Restricted features (IDIAP internal only)

Have a look here.

Last update: September 4, 2006.