7
$\begingroup$

How long does it take ImageIdentify to give an answer for one image? Also, I suppose using it requires an Internet connection with Wolfram servers. Is that correct?

Improve this question
$\endgroup$
2
  • $\begingroup$ It downloads data just like CountryData etc., after that it is quick. A test image of flowers of size 2000x1329 took 0.5 seconds. This question would have been answered quickly in the chat by the way. $\endgroup$
    – C. E. ♦
    Commented May 16, 2015 at 20:59
  • $\begingroup$ The first time I ran it on the Lena test image, it downloaded data for quite a while (maybe close to a minute on a home cable connection). After that, it took 0.088s to identify the image. The result was "person". Also yes, it does need to establish a connection to the Wolfram Knowledge base, whatever that is exactly. $\endgroup$
    – MarcoB
    Commented May 16, 2015 at 21:00

1 Answer 1

Reset to default
14
$\begingroup$

If you got V10.1 you can test it yourself. Make sure ImageIdentify got it's data after say 1st execution and then it can work offline. But use it online from time to time because it is a learning function and it becomes better with time, needs updated data. To understand better the workings, implications and applications of this function take a look also at imageidentify.com project which is based of this function and is a spectacular example of interation of deep neural networks, Wolfram Data Framework (Wolfram|Alpha, Entety) and Wolfram Cloud. I recommend reading the how it works section.

i = Import["https://wolfr.am/4RvEuxmc"]

enter image description here

i // ImageDimensions

{1920, 1080}

imgs = ImageResize[i, #] & /@ Range[200, 2000, 200];

Here we go OFFLINE:

Block[{$AllowInternet = False},
 dat = Flatten[{ByteCount[#], AbsoluteTiming[ImageIdentify[#]]}] & /@ imgs;]

So recognition is scale invariant:

dat // TableForm

enter image description here

And here are the times plotted:

ListLinePlot[dat[[All, ;; 2]], PlotTheme -> "Business", 
 FrameLabel -> {ByteCount, AbsoluteTiming}]

enter image description here

But I think this timings also will improve with time. For the benchmark this was on Mac 2.7 GHz Intel Core i7.

Improve this answer
$\endgroup$
4
  • $\begingroup$ What happens if the result is wrong? How does Wolfram improve their classifier. For instance, I gave it a picture of a European badger and it thought it was an opossum, then an armadillo. $\endgroup$
    – Histograms
    Commented May 18, 2015 at 15:47
  • $\begingroup$ @Histograms "ImageIdentify is being progressively enhanced, especially with training data derived from feedback given by users of this site." - that from the page I linked above and said: "I recommend reading the how it works section." $\endgroup$
    – Vitaliy Kaurov
    Commented May 18, 2015 at 16:17
  • 1
    $\begingroup$ Just FYI: There should be a factor 3 speedup for the upcoming 10.2. $\endgroup$
    – Sebastian
    Commented Jul 2, 2015 at 5:14
  • $\begingroup$ I understand why it has to connect to the internet, but why does it need to connect to an account?! $\endgroup$
    – anderstood
    Commented Feb 7, 2018 at 0:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.