questions about SLN-VIZNAS-IOT

dave_choi · ‎01-04-2021

Hi, everyone.

1. Maximum number of saved faces

In the figure below, the maximum number of faces values are different.
How many faces can you save?

Questions about Anti-spoofing

Is it anti-spoofing (liveness detection) only when using IR+RGB?
Does the Max # faces(IR) in the picture above mean RGB+IR?
Doesn't Max # faces (RGB) support anti-spoofing?

Questions about Light/Heavy Model

What is the standard to distinguish light from heavy?

Thank you for your attention

jeremyzhou · ‎01-05-2021

Hi,
Thank you for your interest in NXP Semiconductor products and for the opportunity to serve you.
1) How many faces can you save?
-- Please refer to the below figure.

2) Is it anti-spoofing (liveness detection) only when using IR+RGB?
-- Yes.
3) Does the Max # faces(IR) in the picture above mean RGB+IR? Doesn't Max # faces (RGB) support anti-spoofing?
-- The anti-spoofing feature is independent of the SLN-VIZNAS-IOT performance, this feature is always enabled during recognizing the face.
4) What is the standard to distinguish light from heavy?
-- In general, the major difference between the light or heavy model is the inference time, the light mode is ideal for use-cases that require fast recognition inference time, the heavy model is aimed at hard-wired/wall-powered use-cases, such as E-lock case.
Have a great day,
TIC

-------------------------------------------------------------------------------
Note:
- If this post answers your question, please click the "Mark Correct" button. Thank you!

- We are following threads for 7 weeks after the last post, later replies are ignored
Please open a new thread and refer to the closed one, if you have a related question at a later point in time.
-------------------------------------------------------------------------------

dave_choi · ‎01-06-2021

hello,

This is an additional question for what you answered.

3) Does the Max # faces(IR) in the picture above mean RGB+IR? Doesn't Max # faces (RGB) support anti-spoofing?
-- The anti-spoofing feature is independent of the SLN-VIZNAS-IOT performance, this feature is always enabled during recognizing the face.

=>Are "anti-spoofing" and "liveness detection" different? Can you briefly explain anti-spoofing?

4) What is the standard to distinguish light from heavy?
-- In general, the major difference between the light or heavy model is the inference time, the light mode is ideal for use-cases that require fast recognition inference time, the heavy model is aimed at hard-wired/wall-powered use-cases, such as E-lock case.

==> Can you explain the following information?
"<500ms (light model <1k faces)
<1000ms (heavy model <3k faces)"
Is the number of faces that can be saved differently depending on the light/heavy model?

Thank for your support!

BR,

Dave

dave_choi · ‎01-06-2021

Hi,

Thanks for your reply.
1) Are "anti-spoofing" and "liveness detection" different? Can you briefly explain anti-spoofing?
-- In general, the anti-spoofing tech is implemented via liveness detection.

=> You said that you can't use "liveness detection" when using only RGB cameras, so can you think that anti-spoofing doesn't work when using only RGB?

2) Is the number of faces that can be saved differently depending on the light/heavy model?
-- The format of save feature data of the registered face is identical in either light or heavy model.

=> Can you store 3000 faces in a light model?
If so, why do you use the Heavy model?
Is it more accurate?

Thank you for your support

BR,

Dave

jeremyzhou · ‎01-06-2021

Hi，
Thanks for your reply.
1) You said that you can't use "liveness detection" when using only RGB cameras, so can you think that anti-spoofing doesn't work when using only RGB?
-- For this question, I've confirmed with the AE team, and they say the liveness algorithm mainly relies on IR camera and partially depends on RGB frame algorithm as well. So we need both dual camera working.
2) Can you store 3000 faces in a light model?
-- No, I don't afraid.
3) Is it more accurate?
-- The main difference between the 'lightweight' and 'heavy' model is capable of recognizing the number level of the face, and this performance is not much related to the inference accuracy of the model.
Hope this is clear.
Have a great day,
TIC

-------------------------------------------------------------------------------
Note:
- If this post answers your question, please click the "Mark Correct" button. Thank you!

- We are following threads for 7 weeks after the last post, later replies are ignored
Please open a new thread and refer to the closed one, if you have a related question at a later point in time.
-------------------------------------------------------------------------------
'

jeremyzhou · ‎01-06-2021

Hi,
Thanks for your reply.
1) Are "anti-spoofing" and "liveness detection" different? Can you briefly explain anti-spoofing?
-- In general, the anti-spoofing tech is implemented via liveness detection.
2) Is the number of faces that can be saved differently depending on the light/heavy model?
-- The format of save feature data of the registered face is identical in either light or heavy model.
Have a great day,
TIC

-------------------------------------------------------------------------------
Note:
- If this post answers your question, please click the "Mark Correct" button. Thank you!

- We are following threads for 7 weeks after the last post, later replies are ignored
Please open a new thread and refer to the closed one, if you have a related question at a later point in time.
-------------------------------------------------------------------------------