NXP Semiconductors SLN-VIZNAS-IOT, Руководство пользователя

Страница: 52 / 64

52
SLN-VIZNAS-IOT User Guide, Rev. 1.1, 10/2020 NXP Semiconductors
3.5 App Configurations
The below table shows a list of the different available app types and their corresponding
feature sets. In this section, each of these app types will be looked at in more depth to
highlight the advantages and disadvantages of each.
App Type
Max # of
Faces
Detection
Speed
Recognition
Speed
Liveness Detection &
Anti-Spoofing
E-Lock (Light) 50 ~80ms ~200 ms Yes
E-Lock
(Heavy)
100 ~80ms ~700 ms Yes
Door Access
(Light)
1000 ~80ms ~200 ms Yes
Door Access
(Heavy)
3000 ~80ms ~700 ms Yes
Userid 1000 ~80ms ~200 ms No
Table 12: App Configurations
As shown in the table above, the various app type configurations greatly affect the
recognition speed, number of faces, and security functionality of the kit. Each app type
has its own advantages and disadvantages.
3.5.1 Light vs. Heavy
As shown in Table 12: App Configurations , E-Lock (Light) is roughly 500ms faster at
recognizing a face when compared to the E-Lock (Heavy) app type. However, the heavy
configuration is capable of recognizing roughly double the number of faces as the “light”
app type. A similar tradeoff between recognition speed and max # of faces can be seen
in the Door Access (Light) vs. Door Access (Heavy) models.
The disparity between the two comes down to the difference in the AI model which is
used to perform the recognition stage of the inferencing pipeline. The light models are
designed to recognize faces more quickly, but are more limited in the total number of
faces that can be recognized, whereas the heavy models are optimized for the max
number of faces that can be recognized, resulting in slower recognition speed than the
light models.
Choosing between the heavy and light model configurations comes down to a question
of performance vs. # of users, and that is generally determined by the use-case.