Technical Specification

This page provides technical metrics of Face AR SDK feature performance. The values below are for your reference only. We received them on fixed conditions. However, many factors may influence the actual technology performance, e.g. the state of device, other apps running, wi-fi enabled, room temperature, etc. We encourage you to test each feature within your environment.

Please visit SDK Features and SDK Features for Unity for the information regarding features support on different platforms.

note

FPS — Frames per second of the face detection algorithm on a given device.
Angles — Maximum angle on which the technology was able to work during the measurement.
Distance — Maximum distance on which the technology was able to work during the measurement.
Real-time (online) — Technology performance in real-time.
Photo (offline) — The processing time needed to take a photo or process it from the gallery.

SDK Features#

Face Detection#

Android

	Android Mid Samsung Galaxy S7	Android Top Pixel 3 XL
FPS	30	30
Angles, degree	70	70
Distance, cm	120	120

iOS

	iOS Mid iPhone 7	iOS Top iPhone 11
FPS	30	60
Angles, degree	70	70
Distance, cm	120	120

Single-face Tracking#

Android

	Android Mid Samsung Galaxy S7	Android Top Pixel 3 XL
FPS	30	30
Angles, degree	70	70
Distance, cm	180	200

iOS

	iOS Mid iPhone 7	iOS Top iPhone 11
FPS	30	60
Angles, degree	70	70
Distance, cm	180	200

Multi-face Tracking#

Android

	Android Mid Samsung Galaxy S7	Android Top Pixel 3 XL
Faces Number	3	3
2 Faces, FPS	20	30
3 Faces, FPS	20	30

iOS

	iOS Mid iPhone 7	iOS Top iPhone 11
Faces Number	3	3
2 Faces, FPS	30	60
3 Faces, FPS	30	60

Faces Number* — Maximum number of faces which SDK can track with acceptable quality and performance on most mobile devices. The actual number allowed for multi-face tracking is not limited by the technology but only by the physical ability of the device and proportions of the screen.

Effect performance#

Banuba SDK allows for a variety of Face AR effects. Some of them require only face tracking and can be represented as a single 3D mask with textures and materials. Other effects are implemented with separately trained neural networks.

Below, you may find information on the real-time performance of Face AR effects which require only face tracking, i.e. face filters, avatars with action units, beautification, and makeup filter (without lipstick).

Android

	Android Mid Samsung Galaxy S7	Android Top Pixel 3 XL
FPS	30	30

iOS

	iOS Mid iPhone 7	iOS Top iPhone 11
FPS	30	60

Beautification#

Beautification filter#

Basic face beautification filter includes skin smoothing, morphing, teeth whitening, eyes flare and LUT. It requires only face tracking, so please, refer to the Effects performance section. Other features given below are based on neural networks, and their performance differs.