Face Detection Performance

Sighthound's face detector performs better than any other publicly available face detection algorithm, including commercial competitors and an open source alternative. Benchmarked using the BAO and Annotated Faces in the Wild (AFW) publicly available data sets containing over 1,700 faces from 600 images, Sighthound's face detector has both the lowest number of false detections and the highest number of correct detections. You can try the face detector in the Sighthound Cloud service.

  True Positive (TP) False Positive (FP) False Negative (FN) Precision = TP/(TP+FP) Recall = TP/(TP+FN)
* Chart shows combined results from Bao Face Database and AFW test set tests ran on Oct. 20, 2016.
Sighthound Cloud 1590 5 184 99.69% 89.63%
Google Cloud 1575 5 199 99.68% 88.78%
Microsoft Research 1370 5 404 99.64% 77.23%
Face++ 1412 6 362 99.58% 79.59%
Kairos 1187 37 587 96.98% 66.91%
OpenCV 1452 278 322 83.93% 81.85%
Definitions
TP: number of identified faces that are correct
FP: number of identified faces that are incorrect
FN: number of missed faces
Precision: percent of identified faces that are correct
Recall: percent of correctly identified faces to the total number of faces