What Kind of Dog Is It – Using TensorFlow on Mobile Device (2016) | Hacker News Reader

What Kind of Dog Is It – Using TensorFlow on Mobile Device (2016)

jeffxtang.github.io

66 pointsputnam22 comments

The Play Store user reviews are priceless. Shows that you can't just put anything and call it AI, it has to work in practice. This model is a nice technical demo for reducing the size of a neural net, but bad at what it says it does.

https://play.google.com/store/apps/details?id=com.ailabby.do...

Mom is 12 different kinds of dogs lmao

Working on my Not Dog app now.

Visiting the Play Store link using an iOS device I can't see any of the reviews.

For example, one star: "My German shepherd came up as 6 different dogs and just for fun I did my mom an my mom came up as 12 different dogs so I will say this don't work at all".

Really turns, "on the internet, no one knows you're a dog" on it's head.

On iOS I can see them- you have to scroll the stars section horizontally.

The comments also hint at the general public's expectations at how smart computers should be. The level of outrage for a free app that purports to do something magical is funny.

Edit: It is free on Android...

Sadly the iOS app isn't free.

https://itunes.apple.com/us/app/dogsnap/id532468586?mt=8 is similar in features and, judging from my experience with it and the comments here on this app, in prediction quality.

I found it very useful in keeping young nephews entertained. It probably also will do well at parties, after a few too many beers.

Just tried What Dog from Microsoft and seems fine. However, depending on the angle it says that my yorkshire is a west highland white terrier :p

You see the same comments here when someone posts an open source library that doesn't do what it should

Well, yes, but the chances that an AI app would be highly accurate at identifying a breed no matter the angle, quality, etc, of the uploaded photo seems obviously low to me.

Coupled with some of the language in the reviews. Not talking about comments like "disappointed" but the more scathing ones.

I don't think I've ever seen anyone here tell an open source contributor "sucks to be you"

I guess it's not very clear that the percentages are confidence in the predictions (right?), rather than genetic percentages (i.e., a cross-breed dog being 50-50% of two breeds, which would be super impressive).

What Dog A Microsoft Garage Project by Microsoft Corporation https://appsto.re/us/OE1M-.i

They're good dogs.

not hotdog.

I love seeing TensorFlow projects like this, very cool.

hot dog

What about hot dogs?

I thought that was the joke

Reply on news.ycombinator.com