Google’s ML Kit makes it easy to add AI smarts to iOS and Android apps | Hacker News Reader

Google’s ML Kit makes it easy to add AI smarts to iOS and Android apps

techcrunch.com

197 pointscoloneltcb23 comments

Can't stop thinking about the centralization going on here, even if you avoid using Google apps now any app could be feeding back your information to Google.

Too bad "Ingredients" labels like they have on food are impractical in the App space. But maybe Apple could highlight key APIs and net connections that it finds in use during app review. Something like, "WARNING: This app makes silent, background connections to [Google, Facebook, etc etc]"

Like In App Purchase warnings, but for background net connections.

I really like this idea, and while it seems like it would be a bit difficult for Apple to implement and the common consumer to understand, it would still further their goal of providing privacy to their users. Plus it would help consumers decide what services they want to give their information and analytics data to, which is pretty powerful.

This is a fabulous idea - and with the review process requires, I don't see this being unreasonable.

Developers will hate it, but I think Apple can take the hit. Google, too, but their review process is something of a joke.

Yeah like the Facebook login button.

... the models are on-device. That's the whole point.

From the article:

"While Google Cloud already offers a number of similar pre-trained and customizable machine learning APIs, those don’t work offline"

"To power the on-device models, ML Kit uses the standard Neural Networks API on Android and its equivalent on Apple’s iOS."

3p libraries won't be allowed to track your information without your consent, please see GDPR.

This looks awesome. Though, I'm disappointed the face tracking has pretty limited landmarks and no positional tracking, I'm sure they'll improve the models in the future. https://firebase.google.com/docs/ml-kit/face-detection-conce...

This landmark tracking on the mobile is new for tensorflow . The benchmark here is dlib which does an implementation of One Millisecond Face Alignment with an Ensemble of Regression Trees paper by Kazemi and Sullivan.

Anyone know comparison benchmarks?

Google are good at choosing names... http://www.elsman.com/mlkit/

Google: ARCore MLkit Apple: ARKit CoreML

Apple have been using (name)core and (name)kit for a lot longer.

As an aside, how do you escape a star character on HN without dropping to a code block?

https://news.ycombinator.com/formatdoc

You need a space after the star, (* core) which is not like escaping at all...

nit: I think the apple naming convention has been (name)kit and core(name).

I am really only familiar with webkit and webcore, so I just assumed most others followed that pattern!

CoreML always sounded wrong to me because of that.

Apple seems to use both, but they've been leaning towards Core* rather than *Core recently. We have CoreAudio, CoreBluetooth, CoreFoundation, CoreGraphics, CoreImage, CoreLocation, CoreText…but there's also WebCore, JavaScriptCore, QuartzCore, and ImageCaptureCore.

https://en.wikipedia.org/wiki/Go!_(programming_language)

Jesus, so even super obscure languages that virtually no one uses is a problem?

Well it is a little late now, but there was certainly some discussion about it when the language was announced.

Why do they provide another API for offline face detection, when Google Play Services already comes with one?

It would be great if apps that use the existing offline face detection were just switched to the better model, without needing to change any app code.

And, if the problem is that some devices can't run the new models, it would better for that to be handled by the API provider, than by each app developer.

Try again, please?

Almost by definition -- one does not simply "add smarts" (of any kind) to an app.

For whose definition of "easy"?

Can we axe the marketing-speak here? It's just released; developers get to decide how easy it is, not Google PR / Techcrunch right?

Would it be a better page if it added footnotes to all adjectives "* For some definition of easy|some-adjective." :P

All joking aside, it's just a click bait title - that's how these things go. Try it out yourself and blog about your experiences with it!

Reply on news.ycombinator.com