Skip to main content

Table 2 ‘Off-the-shelf’ frontal face detector performance

From: A review of image-based automatic facial landmark identification techniques

  BioID (N = 1521) MUCT (N=3755) HELEN# (N = 2330) 300W# (N = 600) Menpo (Front)# (N=6679)
Detector % Acc False P % Acc False P % Acc False P % Acc False P % Acc False P
OpenCV 1 96.45 184 97.95 616 89.06 3548 75.33 6203 73.24 2775
OpenCV 2 96.12 28 98.24 77 84.64 447 71.17 3574 67.35 387
OpenCV 3 96.45 52 98.62 135 85.88 821 72.00 3966 69.35 670
OpenCV 4 N/A N/A N/A N/A 27.21 274 24.83 1597 22.74 349
Dlib 99.34 1 99.89 6 96.82 351 88.17 2488 87.50 339
  1. #Some images in this dataset contain more faces than just the ground truth. These additional faces may be recorded as a false positive result