Startup XNOR.ai, Inventors of YOLO, Video Surveillance Plans Examined

By John Honovich, Published Sep 24, 2019, 11:27am EDT

YOLO is arguably the most discussed and hyped computer vision model today. The inventors of YOLO have a startup named XNOR.ai that aims to bring their expertise to many markets, including video surveillance.

xnor ai profile_2

XNOR.ai is the technology behind the $20 Wyze AI that beat Axis and Hikvision video analytics in IPVM testing.

Inside this note, we examine XNOR.ai's technology, go-to-market approach, how they compare to Ambarella and Intel Myriad plus the potential advantages and problems they face within video surveillance.

Company ********

*** **** ******** ** XNOR.ai ** *** ******* learning *********** *********. *** company ******* *** ** *** Paul ***** ********* ** AI**** **** ** ***** core ******* ****** **** professors ** **** ******** at *** ********** ** Washington (**** ** ****). In ****, ********** ~$** ******* ** funding.

*** ******* ** *********** heavy, **** **** **** 50% ** *** ********* on **** **** *** virtually **** ** *****, according ** ******** *****:

Technology ***** - **********

**** ** ******* ** maximizing *** ********** ** deep ******** ****** ** run ** *** *******, like ** *******. ***** it ** **** ** deliver **** ******** ** academic ********** **** ***** unlimited ********* *** **********, many ****-***** ************ **** video ************, *** ********* and ********** *******, **** AI ********* ** ***-****, low-powered, ***-******** *******.

*** ******* ** ********* XNOR ******* ** ****, a $** ****** **** a ~$* **** **** was *** ******** *** intended ** ** **** learning. **** ** ********* their **** **** *** video ************, ** ** is ** **********.

************** ** "*******-**** **** Learning"***** ** * ******** vision ******* ****** **** runs ****** ** ********, their ***** ******** *****:

* ***** ******* ** for ****** **** * computer ******-***** ****-**** *** of ***** ***** ********:

Turnkey *** *************

**** ** *********** **** to ** ******* *** manufacturers, ** **** **** do *** **** ** have **-***** ** ********* nor **** ** ****** the **** *** **** to ******* **.

** ******* ** **** is ***** '*****' ************* ** **** ****'* build **** *** ******** but *** ** ********. A ************ ******* **** hardware **** *** *****:

**** **** **** **** model ** **** ** AI **** **** ** use, ** *** ******** from *** '**********' ********* show:

*** **** **** *** download *** *******, ***** to '**' ***** *** use ** ***** *******.

*** ******* **** ** confusing ***** *** ***** site **** **** ********* is **** **** ** meant *** **********. *** company ********* ** ** that **** **** **** more ********* ****** **** are *** ****** ** that **** *** ********** use.

Comparison ** ***** - *********, ******, *****

*** **** ***** ***** surveillance ************* **** ***** deep ******** ** **** a ***** ********** ** chips (****) **** ********* GPU ** ***** ******** support *** ** *********.

****.** ** ** *********** to ****. ******* ** having ** *** *** hardware, ****'* ****** *** to *** ** ******** chips / ********.

Advantages - ******* ******** / ***** **** ***

**** ******* *** ********* advantages ** ***** **** - ******* ******** ******* and **** ** ***** cost *** *** ****.

*****, ** **** ************ with ****, **** *** take ******** ******* ** recorders, *** ******** *** deep ******** *** *** AI ** ***** *******. Given *** **** ** millions ** ******* ******* deployed (******* ****, *****, Hanwha, *********, ***.) **** could ******** * *** of ***** ** ***** having ***** *******.

******, ******* ** ***** generally ************* **** ******** SoCs *** *** '** enabled' *******, ************* *** save ***** ** ***** less ********* **** **** paying ****'* ******** ******* (i.e., ** / **** XNOR ******** ***** **** than ******** *****).

Comparable - *************

*** ******* ********** **** the ******** ********** ** video ********* **************** (******** ** ****** in ****). ************* ******** * niche ******** *** **** years. ***** *** ********** was ********* **** *** XNOR *** ******* ********* on *****, ******* ***** remain.

Challenges - ***** *************, ******* *** ******* *** '***'

** *** * **********:

  • ****** ************* **** ****, Dahua, *** ********* ********* want ** **** **** technology ***********, **** ** in-house. ***, ******* **** are *** ****** ** justify *** **** *** two, ********, **** ***********, they ** *** **** to ** ********* ** technology **** ******** **** can ***. **** ***** make ** **** *** XNOR ** *** ***** with ******* **** **** a **************** ***** ** the ***** ******.
  • **** ************* **** ** sell *** *******. ***** the **** ** * free ******** ******* ******** AI ** ******** ******* would **** **** *********** and ***-***** *****, ************* will ****** ** ********* of * **** **** could ******** ********** *** products, ****** ******* ** upgrade ***** ***** ** charged (****** **** * move ***** ** ***** uncommon).
  • **** ************* *** *** to *** *** ******* efficient ****** ** ***** own **** ***** ** close ****** ** **** enough ** **** **** can **. ** ** hard *** ** ** know *** ***** **** could ***, ***** *** lack ** ********* ** AI. ** *******, ** the ***** ****, ** would ** ****, ***** XNOR's **-***** ********* *** that *** ******* **** time.

***** ********** *****, ** believe **** **** ******* manufacturers **** ** *** have *** ********* *** competitive ********** ** ***** larger ******, **** **** XNOR ** ** ** attractive ******.

Poll / ****

Outlook - **** *** ********, ************ *** ****

***** ****'* ******** ** certainly **** *** *** industry ***** ** ******** a ********* ****** *** delivering **, ** *** not **** *** ********** XNOR **** ** ** video ************, ********* ******* of *** ******** ***** challenges ***** ******** ********.

** *** ***** ****, the ******* ** * startup *** ** ****** still ******** *** **-**-****** approach.

******* ** *** ************* use **** ** **** other *****, ** ***** it ** **** *** the ****** *** ***** to **** **** ** improve ** ************ ** video ************ ********.

Comments (4)

I think this will benefit some of the smaller companies that are looking to leverage some of these features. I think this will be big in the DIY/consumer market (Wyze is a good starting example of that). I think if they can prove some value to some of the big guys that are not really hopping on the AI/ML train yet (I'm thinking Bosch/Honeywell/Panasonic/Sony), that would be a huge win for the Xnor crew.

Agree: 1
Disagree
Informative: 1
Unhelpful
Funny

If and when an android based camera comes to market, what stops XNOr from hopping on such a device

this would be a good long play in my opinion

Agree: 1
Disagree
Informative
Unhelpful
Funny

Xnor was one of the AI partners to Bosch spinout SAST at GSX this month, demonstrating exactly this - their AI running on the the SAST (Android) generic camera platform.

Agree
Disagree
Informative: 3
Unhelpful
Funny

Yes, I did not cover this but here is XNOR's blog post detailing that - Adding Deep Learning to your Edge Device with 5 Lines of Code — Xnor and SAST

My perception is that it's good marketing (SAST is well organized and well funded) but selling licenses via 3rd party in the SAST 'store' won't be a big moneymaker (certainly not short term since it's just starting up). If XNOR is going to get significant market share/volume, it has to be by selling to manufacturers with existing sizeable unit sales, that or pivoting to offer their own direct XNOR encoders / cameras / etc.

Agree: 3
Disagree
Informative
Unhelpful
Funny
Read this IPVM report for free.

This article is part of IPVM's 7,327 reports and 971 tests and is only available to subscribers. To get a one-time preview of our work, enter your work email to access the full article.

Already a subscriber? Login here | Join now
Loading Related Reports