Hikvision's ImageNet Win Analyzed

By John Honovich, Published Nov 28, 2016, 01:30pm EST

Hikvision is pushing hard to move up market and win at video analytics.

One notable sign is Hikvision touting [link no longer available] their #1 ranking for an ImageNet challenge.

We hired a computer vision academic, expert at ImageNet, to review the results and assess Hikvision's performance. Inside this note, we explain how ImageNet works, how Hikvision did, the role of GPUs and what this means for Hikvision's pursuit of the video analytics market.

Executive *******

*** ******** ***** ** *** academic's ******:

  • *** *** ************** *** Hikvision *** ***** ******** access ** ****, *** critical ************** ** **** competitions.
  • *********'* ********** / ********* approach *** ***** *** credible *** ****** ******** computer ****** ******** **********.
  • *********'* *** ** ** area **** ***** ********* application ** ***** ************ while ********* *** *** enter *** *** ********* that ******** **** *****.
  • *******, **** *** * solid ********* **** *** Hikvision *** *** *******, by **********, ** *** prove **** ***** ***** technological ************.

#1 ** ***** **************

********* ******** *** *** score ** *** ***** classification *********. *** **** practical *********** ** ****** style ***** ******, *.*., enter ** * **** like '**** ****' ** 'rainforest' *** *** ****** of ***** ******.

*** ********* **** ******** of ****** **** ********** ********, ******** ** ***** are ***** *****:

*********'* ***** ** .**** in ***** ************** ******* translates ** ***** **% accurate ** ******** ** image ** *** ***** 5 *******. **** ***** signal ********* ****** ************ in ***** ************ ********* but ********* ** ***** surveillance *** ***** ** unchanging *** * ***** fixed ****** ** ********* scenes ** **** *********** useful.

#2 ** ****** *********

********* **** **** ** second *** ****** ********* across "*** *****-***** **********" **** * ***** of *.**. *** ****** explained **** ***** ** average **% *********** ** detecting *** ****** ********* (CMU *** ***** ** 66%). *******, ***** ****** per ****** **** ************* as **** ******* *** very **** *** *** system ** ********** ****** (e.g., ****** *** **** given ***** ***** **** and **** ********) ***** others *** **** ****** like ******** (**** ****** size *** **** ******** texture).

No ***** *** ***** ****

** *** * **********, only * ** **** used ***** *****, *** others **** **** ******. Unfortunately, ********* *** *** enter *** '****** ********* **** *****' *********. *** ****** noted **** **** ************* limit ******* *** ** resource ***********. *******, *** video *** *** ********* the **** ******** *** also ********* ****** / lower ********** *****.

GPUs ***

*** ****** ********** **** GPU ************ *** *** key ******. *** *******, they ********* **** ******** students ******* ********** *** access ** **** ****. The **** **** **** have, *** **** ******** that *** ** ****, the ******* *** ******** can ** ********, ******** for **** ******* **********.

*******,********* *** ****** *** ****** to ****** ******* *** ***** *** first ** ***** *** one ** *** ***** globally ** ****** ******'* DGX-1, ***** ****** ******* as '*** ******* ** a ***'. *** **** test, ********* ******** ***** '*****-***** ***-******** *** ********'. ** ********, *** ****** noted **** **** *** top ******** ************ ************* in **** ********** **** limitations ** ********* ****.

**** *******, *** ****** noted **** *** **** difference ** *********** ******* 'poor' ******** *** '***' entrants *** ****** ** *** availability.

*********Hikvision ********

** *** **** ****, the ****** ********** **** Hikvision's *********** (***** ** *** *******) ****** * ********, modern ******** ** ****** and ******** ***** *******. Also,**** ** *********'* ********** deck ** *** ******** results.

Questionable ************* ** ***** ************ ********

*** ****** ********** **** all ************ **** *********** to ******** ***** ******* to *** **** **** sets ***** *** **** is ** *** *** competition. *** ****** **** that **** **** ** broad '****-**********' ***** ***** lead ** ************* ****** results ** **********.

** *** ***** ************, the ****** ********* **** difficulties *** *************** ******** to ********. ** *** positive ****, **** ***** surveillance, *********** *** ** made ***** *** **** and *********** ** ******* (like ****** *** ****) that ***** **** ********* and ************** ******. ** the ********, *** ******** images **** *** ****** resolution, ****** ******* ***** and ******* *** *** light ******** ** ** wash *** ****** ** video ************. ********, ***** *** no *********** ** ******* / ********* ********* **** they ***** ** ** a ********** ***** ************ deployment.

Good ***** ******** / **** **********

*** * ******* **** is **** ***** *************** for *** **** ******* and ***** *********, ***** ranked ***** ***, ** some ***** *****, *** research ************ *** ****** companies ** ************ **** for ********* ***** *****. Moreover, *** *********** ****** from ***** *** ********** will ********* **** ******* developing ********** *********.

Not ****** *** ***** ** ******

** *** ***** ****, the ***** **** ******* directly ************* ** ********** video ************, ***** ***** much **** ***** ****** and ************* *** *****. However, ********* ********* *** the ********* *** ************ plus the ****** ****** ** dominate ***** ********* ** their ******** ****** ** watched *******.

Comments (4)

Good Analysis.

Do you think that Dahua's recent LFW win, is more pertinent then to surveillance video? Maybe your ImageNet expert is qualified to assess that test as well?

Dahua Technology Sets New Record For LFW Facial Recognition

Agree
Disagree
Informative
Unhelpful: 1
Funny

I think it's less pertinent because (1) Hikvision has made much clearer direction of releasing intelligent video globally and (2) intelligent video has much greater overall demand than facial recognition.

In terms of an expert, we would want a different expert who focuses on faces in the wild as the two tests / fields are in the same overall space but different specializations.

Quite bluntly, Dahua struggles to sell $80 HD cameras so I am not bullish on Dahua's chances of delivering a production ready ground breaking facial surveillance system. Hikvision is much more organized so along with 1 and 2 above makes it more of a focus for coverage.

Agree: 1
Disagree
Informative
Unhelpful
Funny

Amended:

Hikvision received the top score in the scene classification challenge. The most practical application is Google style image search, i.e., enter in a term like 'Chinese Government Target #1' or 'Chinese Government Target #2' and get images of those scenes.

Agree
Disagree
Informative
Unhelpful: 1
Funny: 3

Very interesting article! Hiring a computer vision academic for this article was a brilliant idea.

Agree
Disagree
Informative: 1
Unhelpful
Funny
Read this IPVM report for free.

This article is part of IPVM's 7,008 reports, 931 tests and is only available to members. To get a one-time preview of our work, enter your work email to access the full article.

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