Healthy Skepticism for Deep Learning Is Prudent

Author: John Honovich, Published on Jul 26, 2017

The hype for deep learning in video surveillance is accelerating.

Between the race to the bottom and dearth of a 'next big thing', certainly pent up demand exists.

But now is the time for a healthy skepticism about when and what deep learning can do. In this note, we examine the hype and real issues that professionals need to consider.

*** **** *** **** ******** ** ***** ************ ** ************.

******* *** **** ** *** ****** *** ****** ** * 'next *** *****', ********* **** ** ****** ******.

*** *** ** *** **** *** * ******* ********** ***** when *** **** **** ******** *** **. ** **** ****, we ******* *** **** *** **** ****** **** ************* **** to ********.

[***************]

'Revolution' ****

**** *****,*** ******** **** ********** ** "* ********** ** ***** *********", *********:

********************** **** ***** **** ********** ******* **.* ******* ******** ** ************, ***** ************ ******* ***** ******** ** ******* ** *******.

************* **** ***** ***** ***, **** ***** ***** ** *** even *** *** **.* ******* ** ***** ******** *** *** mentioned.

*** **.*% ****** **********.

Irony ******

*** ***** ** *** *** ***** ** *** ******* ******** of *** **** ******** *****:

*** ***** ********* ******** ** ***** ******* *** *** ************ damage ** ********* ** * ****** **previously ********* *** **** and delivering too little in the past. [emphasis added]

*** ****** "***** *** ****** ******** *** **** *** ********* to ****** **" ** ********* *********** ****.

Robots

**** ** ****** ***, '******* *******' ******** ******* *********** ************ *** ****** *** ****** *** **** *** "**** ** robots **** ****** ********** *******". ** *******:

** **** ** ** ** ******** ********* ***** **** **** a **** **** ***

** ********, ** *** **** ******** ***** *** *****. ** are ** ** '******** ********* *****' **** ****** **** ****. Unfortunately, *** *******, ************** **** ****** * ****** ****** ***** ****.

Timing ******

** ** **** ** *** **** ********** *** ************ **** become ********** ** ********* ****** **** ******. *** ******* **** eventually *** *** *** ***** ******.

*** ****** ****** ** ***********.

***** *** / ***** ** ***** ****** ********. *** ********* are ****. *********** *** ******** *****.

Deep ******** **********

**** *** * *** ********* ******* *** * ******* ********** of **** ******** ** *********:

  • ******* ************: **** ******** **** ** **** ** ******* ***** ************ applications. ***** **** **** ****** **** *** ***** **** **** poorly ** ***** *******. ** ** ****** ******* **** *** performance **** **** ************* ****** ************. "**** ********" ** ****** to ** ******** ** ********* *** ***** ** **********.
  • ***** *********** ********: ***** **** ******** ************* ****** **** ********, *** *********** will ** ************* ******** ** **** ** ** *** ************ **. *** '**** *****' ** ******* *** ** ***** applications *** ******** ** ********* *******, ******* *** ***** (*.*., people **** *********, ****** ***********), ** ** ******* **** *********** problems **** ** *****. **** ** *** ******* ** *** easily, ******* ** ******* *** **** ******** ************ *** ** re-trained ** ****** ***** *********.
  • ******* ****** ***********: ****** ******* ***** **** ******** **** ******* **** * few *********, *** ******** *** *********** ** **** ******** ***** surveillance ******* ** ***** ** **** ************* ***** ** *** engineering ***** (***** ** ** ***** ****** *******), ********* ********* and **** ********* ** ****** *** ******* ****** ** ***** on. **** ** ****** ** ****** ** *********** *********** ********* across ******* **** ** *** ** **** *** '**** ********' on *** '***'.
  • ******** ***** ******: ****** ***********,*** **** ******** ******** **** *********** ******** ***** ******.

Early *******

**** *** *** *****, ** ** ********* ** ****** ********* that **** ** ***** ***** **** *** **** ******** ** video ************.

* *** ** **** ** ***** ***** ** ***** ********* demos. *** *******, ** *** ****** *** **** **** *****, one **** ******** ******** *** ******* *** *** ******** / performance ** ***** ******. ***** * ****** ** ** ** watching **, *** ***** ****** **** *** **** ****** **** on ******* **. ** ****** ***, **** **** ****** ******* in * **** ** *** *** ******. ***** ***** ****** will **** ********** **** *** *** *** * *** ** deep ******** ** **** *****.

Healthy **********

**** **** ********, ** **** **** * ****** ** ***** for **** ******** ** ****** (**** *********** *** ******** ****) for ** ** **** * *** ********** ** ***** ************. Hyping ** *****, ********** ***** *** ***-**** ******* **** ***** analytic ******** ** *** *****, ***** ********* ******** **** ********'* reputation *** *** ********** ********* ****** **** ** **.

Comments (11)

*** ** *** *** ************ ******** ****** **** **** ********* the ****** *** ******** ** '**** ********' ** ************/******** ************??

*.*. *** ** ********* ** ********?

************ ******** ** * **** ********.

*** *******, * ***** ****** '********' ** *** ******* ** a '********' ***** ** ******** *** **** ****** ***** *** it.

* ** *** **** *** ****** *** ** **** *** around *** **** ********* ******* *** ******** (*.*., ** **** learning ***** ************ '*******' ** **/**'* *******?). @*****, *** **** you ****.

*.*. *** ** ********* ** ********?

***, ******** ** **** ***** ** ***** *********** ********** (*.*.:*.*** ********* **** ******** / *** *********), ******'* ******** *********, ***. ****** **** *** ********* *** defined ** "**********" ************.

**** ** *** ********* ******* ******* ****** ***** (*********, ***.) applied ** ********** *******. **** ******** ** *********, ** ***** so ***, *** **** ******* ******* ** *** ********** *******, doing * ****** *** ** *********** **** ** "*****" (*******, boat, ***, ***) ** *** *****. ***** *** ******* ********* applications ** ****, *** ***** *** *** *********** ***** ******** unique (*** ****** ****** ********* *********** ******).

*** ***** ****** * ******** **** ******** ******* **************, *** it ***** *** ** * **** ****** ******. **** ** very ******* ********** ** ********* * ***** ***** ********** ***. You ***** ** **, *** ***** *** ** **** **** to ***** * ********** *** **** ******* ***** ***** * competing *** ******* ********** **** ****** ** ***. ******** ******** used ** ***** ***'* *** *** **********, *** ***** *** some ***** ******** ** **** ***** ******** *** ********** *** (many *** ********* *** ***-**********/******** ********).

*** ***** ****** * **** ****, ****-******** ******* *** ******* to ******* **** ** ******* ********* *** ******** ******** ** a ****** ** *******, *** *** ***** *** ***** ******* to *** ** (** *** *** * ******* ***). ****, if ******* **** *** ***** **** ******, *** *** *** tagging ** ***** ***, *** ***** *** ******* * ******* fee (**** ** ******** *** ****** ********** *** *** *****/*********** by ***).

**** ******** ** ***** ******* ** ** **** ***** ** so **** ******** **** ** ** ******** **** ******** ******* will **** ***** ************ ******* ** **** ******** **** ****** the ******** ********.

***** *** ***** *** **** ****** ******** ** ******** ** in ***-*** ********/************ ***** ** *** ** ************ **** **** learning ***** ********. *** *******, *** ***** ** **** ** patent * *** *****-****** ******* ****-******** ******* ** * *** that ** ********* ******** *** ** * *** ****** ********* a ******, ** ******** * ********** ******** ******* ********.

**** ** * ***** *******... ************:

"*** ****** ****** ********* *********** ******"

** ** ******* ************* ** '**** ********' - *** **** me ** ** * ** ***** - ** ******* ** me ** ** **** ** *** ******** ******* ***** ** pretty ****** '**/****' ********* ***** ************.

**** *** *********** *** '**** ********' ********** ** * ******** increase ** ************* ***** *** *****.

*** ****** ***** **/**** ********** *** ** ******, *** **** operational ************ **** ******* **** ***** '**** ********' ** '****' things **** **** *** ******** ******* *** ************* ***** ******** to ***** **** ****.

** * *****?

*** **/**** ******* ** *** * *** *******, *** ** is **** *** * ******* *******. * ** *** **** if *** *** ***** ******* ***** *******/************* ** ***-***** ************, which ***** ****** **** *** *** *** **** ** **** the **/**** *******.

**** ******** ** *** ** (********** ************), ** ** **** about ******* ****** **** **** *** ********, *** **** ***** pattern *********/***********, ************ ** ***** ***** ****** ***** *** ** as **** ** ***** **** **** *** ****.

*** *** ***** * *** ** ****** ***** ***** ** cancer, *** **** **** *** **** ** *** ***** ** know *** ** ********* ** ********* ****** ********* ********.

******* - *'* *** ** ******** *** *** ********* ** not ***** ******....

***** *** * ******** ******* ***** ****** ** **** (***** vs. *** ****) **** ***** *** *** *** ****** **** an *********, ** **** ********. ** **** ** **** ** hardware. **** ** ********* * *** ****, *** **** ***** waves ******* **** ******** ***** **********.

****** ******** ** **** ******** *******, **** ********, **** ******* this ****** ** ********* *** ******** ** ******** ******* ** not ********.

*********, ** ************* ** **** *** *** *** ****** **** Learning **********, *** ****** **** **** ** **** ** ******** like ********** *** ******* ** * *** **** ** *********** to *** *********.

*****...

"* ** *** **** ** *** *** ***** ******* ***** patents/patentability ** ***-***** ************, ***** ***** ****** **** *** *** you **** ** **** *** **/**** *******."

**, ******** * ** ** ********* - * ***'* ***** that ********* *********** ****** ** ********** ******. *** '*****-******' **** set *** ** ********/****** ** ****** *********** ******. **'* **** simple ****.

**** ***** *** ******** ** ********* (* *****) ** *** ability ** **** '*****' **** *** ******* ***** **** **** fast ****** ** ****** * ********* *** **** *** ****** that *****.

** **** ****** *** ** *** ***** *****, *** ********* in *** ******** **** ** ** ********* *** **** **** they **** **** *** ***** ******* - ******** **** **** word **** ********** **** *******.

********* ************* ***** *********** ** ******** ** *********** ***** ****** on * **** **** ********** ** **** ** ******* **** really **** ******.... *** ** **** ******** ** ****-***** (** VCA ***) ** * ******* ****** **** ***** *** *** capabilities ** ******** ******* ******** ******** ********, *** **** ******** has *** ********* ** ***** ****** ** *** **** - like *** ********* ********** *** ****** **** **** *** ************ to ******** ******** *****.

** ** * ******* *****, *** ** *** ** *** early ******.

***** **** ***** ********* ******** ********* (** ******** ******) *** in ***** ******** ****** **************. **** ******** ***** ******* ***-****** approaches ** *** ** * ******** ****** ** ********** ***** from ***-***** (***** **** *****, **** ******** ********, ******* ****-******* coupled **** ******** ****** **** ** ********* ***** *******, ***.)

** ***-***** ********** *** ****** ******** **** ** **** ** object**, **** ***** ****** *** ** ***** ** *** ** that ** * *******/******/**** ******** ******.

** **** ****, *** ***** ** *** ******** ***** ** analytics. ****** *****, ** *** ****** ** ***** ******* **** an ******** ********* ******, ** ******* **** *******, *******, *** expensive. ***** ****** * ***** * **** ** ******* *** embedded *******, ******* ********* *********** ** *** ******/************* **** **** than *** ** (*** ** ***** ******* **** ****-****** **** of ***** ******** ** ***** ****** ******** ****** ********* ***** get *****).

******** *** ** ********* ***** *** *** ******** **** * major ******, *** *** ***** **** **** ****** ********** (*** marketing/sales) *** *** ********** ***** **** ***** ***** *** **** figured *** **** ** ****** ** ** ***** **.

* ***** ************* ************* ** ********* ** **** ** *** ** ******** *** security ********.

*** *** ****** ********* ** **** *** *** **** ** research ** *** *********.****://***.**********.**/.

*********, **** ** **** ** * *********** **** **** ** actual ***, ************ *** ******** ************. **** ***** ****** ** DNN-based ********* ****** *** **** ********** *** **********, ** ********* below.

* ***'*/***'* ******* ** *** "**** * *****" ******** ******* I ** *** **** ****** ***** *** ********** ******* ********* work ** * *** *****.

*** ******** ************, *** *********** ****** ***** ********* ** **** form ** ******* **********, ** **** * ***** ****** (**** likely * ******) ***** ***** ********* ** *** ******. **** could ** ** **** *** ***** ******** ** "***", ** to ****** ***** ********* ********** (******** * ****** ******** ** be ********* **. **** ***-***********).

***** *** * *** ****** **** **** ** *********/**** ********* to ** ** *********** ****** *** ************:

  1. ******** ********* *** ********* *** **** ******/************/****** ******** *** ******* around **. ************* ******* ***** ********** **-*****, **** ****** *****. The ******* ******* ****** ********, ***** **** ** ***** ******** cannot ** ****** ****** ****** **** ***** ** ****** ** attackers ** *** ****** ** * ****** *** *&*/******* ********. Not ****** ** ** **********, *** ** **** **** ****** a ****** ****** *** *** ******* ** **** ******* **** of ***** ********.

  2. ****** ********* **** ******* ******* ** ** ****** ****** ******** giving ** * **************. ******** ******* **** *** ******* ** generally **** ****** ***** ****** *******. ****, ******** ********* ** image ** *************** *********, ** ** ***** ** ***-****** ****** out ** *** ***** *** **** ******* ******** **** **** some **** ********. **** ** ********* * ******* **: **** for **** ******->******* *** ***** ****/*****/****** ***** ********->******* *** "*********** motion"->then ******* *** *** **** ******* *** *** *** **************. This ***** *** *** **** ** ** **** **** **** inject ******** ****** ****** **** *** ****** ** ****** **.

  3. ***** ****** ******* ***********, ** ********** ** ****** *** *** would ****** *** **** ** *******, ****** **** **** ******* with *** **** ******* (********, *** ****** ** ** ******** more ****** ** *** ****** ** ***** ****** ** *******/********* labeled **** **** ********), *** ******* *** ****.

*********, ** **** * ******* ****** ********-***** *** ********* ******, you ***** **** ** ********* **** **** ** ******* **** that ******** ********** *******, **** ******** ****** *** ***** ** motion (***** ***** ******* ******** ********** ** **** ******). ** the ****** ***** ******** ******** *** ********* ****** *** **** 1 ******, ** ***** ******** ** *****.

***** ** ******** ***********, ** ** **** **** ** **** which, ** ***, ********* *** ** *** (* ***** **** is **** * ****** ********** *** ******** *** ***** ***********-******** cameras). ******** *** * ********** ****** **** **** ****** ***** require **** ****** ** ****** ********* *** * **** ****** motivated ********. ** ******* ***** *** *** ***** ** ************ for ***** ***** *********, ** *** **** ** *** ****** today, *** ******** *** *********** *** *** ***** ** ****, those ********** *** ******* *** **** **** *** *****-****** *****-*****-***-*********-***** armed ******.

*** **** ** * *********** ********, *** ********* **** ******* would ****** *** ******** ******** *** ** ** ****.

* ***** ***** ******** ** ******** ******* *** ***** ***-******* existing ****** **** ** ******** *** ******** ***** *** ****** layer *** ***** *****. ***** **** ** *** ****** ******** and ******** ***** *** ******.

** * **** ** *****: ****** **** *** **** ** optimize *** **** ** *******. ****** ****** **** *** ******** to ****** *********** ************; ****** ******* ** ***** *** ***** distribution *** *** ********* ** *********** ********. ********* *** *** resistant ** *********** ********.

******* ************** ** ***** **** ** ********* ** **** ******** and ******* ***** ******** ****'* **** ** ***** **** ** optimise *********** ******, ********* ******** ** ******** ** ********* *** working *** **** ****. (**** *** ***** ** **** *****: https://arxiv.org/pdf/1412.6572.pdf)

** **** ********* *********** **** ****** **** ***-*******, ****-***** ****** model ********** ***** ***** ********** ** * ******* **** **** some ***** ** * *****.

** ** ****, **** (********** *********** ********) ** ** ****** area ** ******** *** ** ***** * **** ******** ********* and *** ******* ** ****** ******.

Login to read this IPVM report.
Why do I need to log in?
IPVM conducts unique testing and research funded by member's payments enabling us to offer the most independent, accurate and in-depth information.

Related Reports

Axis Perimeter Defender Video Analytics Tested on Jul 12, 2018
Axis 'high security' video analytics offering is Perimeter Defender, OEMed / developed with Digital Barriers. But how good is Perimeter Defender?...
Hikvision Covers Up Racial Profiling And AI Error on Jun 25, 2018
Faced with global scrutiny, led by the US government-funded Voice of America (VOA), Hikvision has covered up evidence showing their racial...
Axis Guard Suite Video Analytics Tested on Jun 25, 2018
In 2015, we declared Axis' Guard Suite analytics "weak", with missed detections and false alerts common. But after nearly 3 years and a 2.0 version...
Huawei Software Defined Cameras on Jun 25, 2018
Huawei is aiming to break the reputation of Chinese companies not being good at software. The company is now leading their video surveillance...
IFSEC 2018 Final Show Report on Jun 20, 2018
IPVM attended the IFSEC show for the first time this year. The Chinese took over the show, centered on Hikvision, flanked by Dahua, Huawei and a...
Hikvision Corrects False Cybersecurity Announcement on Jun 18, 2018
Hikvision has corrected a false cybersecurity announcement that claimed a British government-sponsored program endorsed the cybersecurity of...
The Dumb Ones: PSA's Bozeman On Cybersecurity on Jun 15, 2018
The smart ones are the hundred people who flew to Denver and spent $500+ on a 1.5-day conference featuring Dahua as a 'cyber responsible partner',...
ReconaSense - The AI / Access Control / Analytics / IoT / Video Company Profile on Jun 12, 2018
One company's ISC West booth stood out for displaying a light-up tower of buzzwords. The company, ReconaSense, pledged to be 'making sense of it...
Bosch IVA Video Analytics And Motion+ VMD Tested on Jun 06, 2018
Bosch's video analytics now ship on nearly every model, from indoor domes to high-end 5MP starlight cameras.  In this test, we evaluate Bosch's...
Dahua's Terrible Cybersecurity, Buys Credibility From PSA And SIA on Jun 04, 2018
Dahua has a terrible cybersecurity track record. But American organizations, like the Security Industry Association (SIA) and the PSA Security...

Most Recent Industry Reports

Security Sales Course Summer 2018 on Jul 13, 2018
Based on member's interest, IPVM is offering a security sales course this summer. Register Now - IPVM Security Sales Course Summer 2018 This...
US Tariffs Hit China Video Surveillance on Jul 13, 2018
Chinese video surveillance products avoided tariffs for the first two rounds. Now, in the third round, many video surveillance products will be...
Last Chance - July 2018 IP Networking Course on Jul 12, 2018
Registration ends today, Thursday. Register now. This is the only networking course designed specifically for video surveillance...
4 Most Difficult Camera Installs (Statistics) on Jul 12, 2018
Heavy housings, cumbersome brackets, heavy ladders required, and tricky field of view requirements will cause difficulties no matter the camera...
Axis Perimeter Defender Video Analytics Tested on Jul 12, 2018
Axis 'high security' video analytics offering is Perimeter Defender, OEMed / developed with Digital Barriers. But how good is Perimeter Defender?...
Hikvision Fights Ban - Claims 'Red Scare', Hires 14 Term Ex-Congressman on Jul 11, 2018
Hikvision is fighting back against the House Bill Ban of their products. Hikvision has hired one of the biggest lobbying firms, led by a 14 term...
Arecont Acquisition By Costar on Jul 11, 2018
Arecont Vision acquisition by Costar Technologies has been approved by the court, concluding the bankruptcy process triggered by Arecont's...
Amazon Ring Partners With Rapid Response For $10 Monitoring on Jul 10, 2018
Amazon's Ring alarm system is using Rapid Response for monitoring, IPVM has confirmed in our testing. Amazon is arguably the most feared new...
SIA Lobbyists Working On House Bill Ban of Dahua and Hikvision on Jul 10, 2018
While SIA is most known for ISC West, SIA maintains the industry's most significant lobbying organization to influence US government action. Last...
Eastern and SavvyTech Merge, Form ENS, Targets ADI on Jul 09, 2018
ADI, ENS is coming for you. Or, at least, they hope. Two US distributors, NY based EasternCCTV and California based SavvyTech have merged, to...

The world's leading video surveillance information source, IPVM provides the best reporting, testing and training for 10,000+ members globally. Dedicated to independent and objective information, we uniquely refuse any and all advertisements, sponsorship and consulting from manufacturers.

About | FAQ | Contact