Technical Specifications

Face Detection

• Robust frontal face detection
• Detection of multiple faces in a photo
• Head rotation support:  –30..30 degrees of in-plane rotation and –30..30 degrees out-of-plane rotation
• Determines in-plane face rotation angle
• Detection speed:
  • Realtime detection (webcam resolution, –15..15 degrees of in-plane head rotation): 0.00154 seconds (649 FPS) (AMD^*), 0.00863 seconds (116 FPS) (iOS^*), 0.01414 seconds (71 FPS) (Android^*)
  • Reliable detection (digital camera resolution, –30..30 degrees of in-plane head rotation): 0.0081 seconds (AMD), 0.05 seconds (iOS), 0.082 seconds (Android)
• Returned information for each detected face: (x,y) coordinates of face center, face width and rotation angle
• Easy configuration of face detection parameters

Face Matching

• Matching of two faces at given FAR (False Acceptance Rate) and FRR (False Rejection Rate)
• Enrollment time:
  • Webcam resolution, using FSDK_GetFaceTemplate: 0.01417 seconds (71 FPS) (AMD), 0.03725 seconds (27 FPS) (iOS), 0.0777 seconds (13 FPS) (Android)
• Template Size: 1040 bytes
• Matching speed:
  • Single thread, templates per second: 5000000 (AMD), 3205128 (iOS), 242600 (Android)
  • Multiple parallel threads, templates per second: 53763440 (AMD), 10101010 (iOS), 1096791 (Android)
• Returned information: facial similarity level
• ROC Diagram

Live Video Recognition with Tracker API

• Assigns a unique ID to each subject detected in video
• Allows tagging any subject in video with a name, and recognizing it further
• No requirement for a subject to pose to be enrolled
• Constant learning of subjects’ appearance
• Provides with estimates of false acceptance rate and recognition rate
• Tracks multiple faces and their facial features
• Recognizes male and female genders
• Recognizes age
• Recognizes facial expressions

Facial Feature Detection

• Detection of 70 facial feature points (eyes, eyebrows, mouth, nose, face contour)
• Detection time (using FSDK_DetectFacialFeaturesInRegion, not including face detection stage): 0.00027 seconds (3733 FPS) (AMD), 0.0003 seconds (3333 FPS) (iOS), 0.00188 seconds (531 FPS) (Android)
• Allowed head rotation: –30..30 degrees of in-plane rotation, –20..20 degrees out-of-plane rotation
• Returned information: array of 70 (x,y) coordinates of each facial feature point

Eye Centers Detection

• Detection of eye centers only, detection time (not including face detection stage): 0.00027 seconds (3752 FPS) (AMD), 0.00028 seconds (3571 FPS) (iOS), 0.00188 seconds (531 FPS) (Android)
• Returned information: two (x,y) coordinates of left eye center and right eye center

Gender Recognition

• Recognition of different genders
• Gender recognition time (not including face and facial feature detection stages): 0.0039 seconds (AMD), 0.0063 seconds (iOS), 0.0122 seconds (Android)
• Returned information: confidence level in each gender

Age Recognition

• Recognition of age
• Age recognition time (not including face and facial feature detection stages): 0.0051 seconds (AMD), 0.0075 seconds (iOS), 0.0131 seconds (Android)
• Returned information: age of a person
• Depending on source quality and lighting conditions, error rate is +/- 5 years

Expression Recognition

• Recognizes if the subject smiles and if the eyes are open or closed
• Expression recognition time (not including face and facial feature detection stages): 0.0043 seconds (AMD), 0.0063 seconds (iOS), 0.0122 seconds (Android)
• Returned information: confidence level in each expression

Liveness detection

• Detects whether the subject is live (i.e. not a photo/video presented to the camera)
• Works with still images and videos
• Liveness detection time (not including face and facial feature detection stages): 0.017 seconds (AMD), 0.016 seconds (iOS), 0.034 seconds (Android)
• Returned information: the probability of the subject being live

Multi-Core Support

• The library uses all available processor cores when executing face detection or recognition functions, to maximize the performance.

Library Size

• The size of the redistributables does not exceed 160 MB for each platform.

^* Measured on AMD Ryzen 5 1600X processor with 12 threads, iPhone X with 6 threads, Google Pixel 2 (Snapdragon 835) with 8 threads.

Download Evaluation Version     Ordering Information