How is Google Lens supposed to improve if you can't provide feedback?
I've been playing around with Google Lens for a few days since I got it on my Pixel. It's far from blown my mind, but one thing I've found really confusing is that the entirety of it is just a button you press and it spits out its best guess (which a lot of the time is nothing).
I'm not machine learning expert, but isn't the idea behind it that you train it by telling it when it gets something wrong, and telling it what the answer should have been? For example, I have a dog that's half Shih Tzu half Bichon Frise. Any time I use Lens on a photo of him it thinks he's a Schnoodle. Pretty impressive guess, but not totally right. At this point shouldn't I be able to correct it so it can take that into consideration for future "guesses"?
It's a similar thing with Google Photos face detection. It's generally pretty accurate, but sometimes it gets faces wrong, or misses a face entirely and there's no way to correct it or manually tag a face. How are these products supposed to ever improve if nobody's telling it when it gets stuff wrong?
Submitted October 30, 2017 at 01:01AM by Ashanmaril http://ift.tt/2ydP2B3 http://ift.tt/eA8V8J
Comments
Post a Comment