Trial speeds vs Paid components

General information about Neurotechnology and its software

Trial speeds vs Paid components

Postby omarmercado8 » Jan 10, 2019 16:31

Hi,

I have been making some tests with the trial versions of Megamatcher SDK.
I have seen very low speeds on executing the Enroll operation.

Im evaluating this product response speeds to know if its suitable for our project so I wanted to ask if the paid products will
offer better enrollment speeds and in general?
omarmercado8
 
Posts: 4
Joined: Sep 24, 2018 17:39

Re: Trial speeds vs Paid components

Postby MartynasV » Jan 11, 2019 07:58

omarmercado8 wrote:Hi,

I have been making some tests with the trial versions of Megamatcher SDK.
I have seen very low speeds on executing the Enroll operation.

Im evaluating this product response speeds to know if its suitable for our project so I wanted to ask if the paid products will
offer better enrollment speeds and in general?

Hello,

Speeds depend on what license you are using and you obtain license in code with NLicense.Obtain() method. Here are speeds: https://www.neurotechnology.com/megamat ... tions.html.

I have seen very low speeds on executing the Enroll operation.

Which version of SDK you are using?
Which method's performance you are testing?
Which modality you are using(face,finger, etc.)?
How many biometrics you enroll in one time? (1 face, 5 fingers, etc.)
How long does it take? (in miliseconds)
Martynas V
Neurotechnology
MartynasV
Neurotechnology
 
Posts: 515
Joined: Feb 27, 2015 11:24
Location: Vilnius, Lithuania

Re: Trial speeds vs Paid components

Postby omarmercado8 » Jan 11, 2019 19:55

Hi,

Which version of SDK you are using?


Im using the MM 10 sdk trial version


Which method's performance you are testing?


NBiometricTask task = biometricClient
.createTask(EnumSet.of(NBiometricOperation.SEGMENT, NBiometricOperation.CREATE_TEMPLATE,NBiometricOperation.ENROLL), subject);

Which modality you are using(face,finger, etc.)?


Finger only

How many biometrics you enroll in one time? (1 face, 5 fingers, etc.)


2 fingers

How long does it take? (in miliseconds)


it takes around 2 minutes for 100 enrollments.

Im saying is slow because I remember reading somewhere in the documentation it could do above a 1000 per minute. Im just wondering if a new license like the MMA extended is bought
will the same code just starts working faster oe new code will be needed?

Thanks for the quick response
Regards
omarmercado8
 
Posts: 4
Joined: Sep 24, 2018 17:39

Re: Trial speeds vs Paid components

Postby MartynasV » Jan 14, 2019 14:41

omarmercado8 wrote:Hi,

Which version of SDK you are using?


Im using the MM 10 sdk trial version


Which method's performance you are testing?


NBiometricTask task = biometricClient
.createTask(EnumSet.of(NBiometricOperation.SEGMENT, NBiometricOperation.CREATE_TEMPLATE,NBiometricOperation.ENROLL), subject);

Which modality you are using(face,finger, etc.)?


Finger only

How many biometrics you enroll in one time? (1 face, 5 fingers, etc.)


2 fingers

How long does it take? (in miliseconds)


it takes around 2 minutes for 100 enrollments.

Im saying is slow because I remember reading somewhere in the documentation it could do above a 1000 per minute. Im just wondering if a new license like the MMA extended is bought
will the same code just starts working faster oe new code will be needed?

Thanks for the quick response
Regards

Hello,

Lest calculate: 2 minutes for 100 enrollments(2 fingers for single enrollment) => 120 seconds for 200 fingers => 0.6 second per 1 finger, you can achieve this speed with Finger Client license. If you want to achieve even faster speeds, then obtain instead Finger Fast Extractor license(a.k.a Fingerprint Image Processing):
Code: Select all
var obtained = NLicense.Obtain("/local", 5000, "FingerFastExtractor");

Then test speeds again.
You should be able to achive up to 3000 extractions per minute if you have powerful CPU as mentioned in our site: https://www.neurotechnology.com/megamat ... tions.html (look for "Fingerprint Image Processing").

MMA(MegaMatcher Accelerator) is different product, it is server side matching software(https://www.neurotechnology.com/megamat ... rator.html), it can't perform extractions.

What extraction speed you need to achieve(fingers per second)?
Martynas V
Neurotechnology
MartynasV
Neurotechnology
 
Posts: 515
Joined: Feb 27, 2015 11:24
Location: Vilnius, Lithuania

Re: Trial speeds vs Paid components

Postby jacobg » Aug 12, 2019 16:38

I'm also having this issue doing face extraction. I have VeriLook licensing, but registered for SDK trial to test MegaMatcher speeds. I call
Code: Select all
NLicense.obtain("/local", "5000", "FaceClient,FaceFastExtractor,FaceMatcher")

But face extraction on single images using NBiometricClient.createTemplate is not only slow, but gets slower each successive call. Here is standard output of the time it takes to call createTemplate 10 times:
Code: Select all
12:28:06.670 [pool-1-thread-1-ScalaTest-running-NeurotecSpec] INFO NeurotecClient - Operation 'Create NBiometricClient' took 61 milliseconds
12:28:07.865 [pool-1-thread-1-ScalaTest-running-NeurotecSpec] INFO NeurotecSpec - Operation 'Create template for subject' took 640 milliseconds
12:28:08.179 [pool-1-thread-1-ScalaTest-running-NeurotecSpec] INFO NeurotecSpec - Operation 'Create template for subject' took 309 milliseconds
12:28:11.224 [pool-1-thread-1-ScalaTest-running-NeurotecSpec] INFO NeurotecSpec - Operation 'Create template for subject' took 3044 milliseconds
12:28:14.506 [pool-1-thread-1-ScalaTest-running-NeurotecSpec] INFO NeurotecSpec - Operation 'Create template for subject' took 3282 milliseconds
12:28:17.829 [pool-1-thread-1-ScalaTest-running-NeurotecSpec] INFO NeurotecSpec - Operation 'Create template for subject' took 3322 milliseconds
12:28:21.160 [pool-1-thread-1-ScalaTest-running-NeurotecSpec] INFO NeurotecSpec - Operation 'Create template for subject' took 3331 milliseconds
12:28:24.533 [pool-1-thread-1-ScalaTest-running-NeurotecSpec] INFO NeurotecSpec - Operation 'Create template for subject' took 3372 milliseconds
12:28:27.832 [pool-1-thread-1-ScalaTest-running-NeurotecSpec] INFO NeurotecSpec - Operation 'Create template for subject' took 3298 milliseconds
12:28:31.160 [pool-1-thread-1-ScalaTest-running-NeurotecSpec] INFO NeurotecSpec - Operation 'Create template for subject' took 3327 milliseconds
12:28:34.556 [pool-1-thread-1-ScalaTest-running-NeurotecSpec] INFO NeurotecSpec - Operation 'Create template for subject' took 3396 milliseconds


Whereas if I use my development dongle license with the products "FaceClient,FaceMatcher", then I get better performance:
Code: Select all
12:51:59.212 [ScalaTest-run-running-AsyncNeurotecSpec] INFO AsyncNeurotecSpec - Operation 'Create template for subject' took 7888 milliseconds
12:51:59.699 [ScalaTest-run-running-AsyncNeurotecSpec] INFO AsyncNeurotecSpec - Operation 'Create template for subject' took 479 milliseconds
12:52:00.196 [ScalaTest-run-running-AsyncNeurotecSpec] INFO AsyncNeurotecSpec - Operation 'Create template for subject' took 491 milliseconds
12:52:00.697 [ScalaTest-run-running-AsyncNeurotecSpec] INFO AsyncNeurotecSpec - Operation 'Create template for subject' took 497 milliseconds
12:52:01.205 [ScalaTest-run-running-AsyncNeurotecSpec] INFO AsyncNeurotecSpec - Operation 'Create template for subject' took 504 milliseconds
12:52:01.696 [ScalaTest-run-running-AsyncNeurotecSpec] INFO AsyncNeurotecSpec - Operation 'Create template for subject' took 487 milliseconds
12:52:02.198 [ScalaTest-run-running-AsyncNeurotecSpec] INFO AsyncNeurotecSpec - Operation 'Create template for subject' took 499 milliseconds
12:52:02.697 [ScalaTest-run-running-AsyncNeurotecSpec] INFO AsyncNeurotecSpec - Operation 'Create template for subject' took 495 milliseconds
12:52:03.203 [ScalaTest-run-running-AsyncNeurotecSpec] INFO AsyncNeurotecSpec - Operation 'Create template for subject' took 502 milliseconds
12:52:03.703 [ScalaTest-run-running-AsyncNeurotecSpec] INFO AsyncNeurotecSpec - Operation 'Create template for subject' took 496 milliseconds


And actually it's even slow using "FaceClient,FaceMatcher" with SDK trial. So there seems to be an issue with SDK trial performance.

Here is the unit test source code:
Code: Select all
  "test repeated template extraction" in {
    implicit val client = NeurotecClient.create

    val image = imageFromResource("/references/2/1.jpg")

    (1 to 10).foreach(_ => {
      val subject = new NSubject
      val face = new NFace
      face.setImage(image)
      subject.getFaces.add(face)
      HelperUtils.measure(logger.info(_))("Create template for subject",
        client.createTemplate(subject)) shouldEqual NBiometricStatus.OK
      face.dispose()
      subject.dispose()
    })

    image.dispose()
    client.dispose()
  }


Did you figure out your issue why the trial is slow?
jacobg
 
Posts: 22
Joined: Jun 14, 2012 18:47

Re: Trial speeds vs Paid components

Postby jbfan » Aug 30, 2019 16:22

Hello,

I had this same issue with SDK Trial Megamatcher 11.1

Even using diferents licenses (FaceExtractor vs FaceFastExtractor) the performace of face extraction continues the same, slower than engine specification in case of FastFaceExtractor.

So, is this because it is a Trial SDK?
The FaceImageProcessing in especifications is the same license of FaceFastExtractor?

Thank you,
jbfan
 
Posts: 4
Joined: Aug 27, 2019 20:22

Re: Trial speeds vs Paid components

Postby jacobg » Sep 03, 2019 10:40

Contact Support. There's a bug with their trial SDK license not giving the fast speeds. I assume they'll fix it in their next version, but they'll give you a 30 day serial number license instead to test with that has the fast speed.
jacobg
 
Posts: 22
Joined: Jun 14, 2012 18:47


Return to Public

Who is online

Users browsing this forum: piterhavy, pitertale and 2 guests