Result: there is a *fixed*, human-invisible map you can add to *any* image, and it renders it unclassifiable by multiple DL systemspic.twitter.com/s4zMsPy6Jy
You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
Result: there is a *fixed*, human-invisible map you can add to *any* image, and it renders it unclassifiable by multiple DL systemspic.twitter.com/s4zMsPy6Jy
I have long suspected that DL image classifiers depend mainly on texture and maybe color, making much less use of shape than vertebrates do…
Yeah, I kind of collapsed two different results there, for concision, but may be misleading.
If the “mostly texture” theory is true, this suggests that different nets look for somewhat different textons
oh god lolling uncontrollably
It’s 3D shape that’s hard… I’ve added some tweets about that
Thanks!
Yes, I saw that too… there’s possible confounds with that one. Would like to dig deeper :)
Trying it with my facebook profile: so far it's a very visible 12% opacity, but fb still knows person, eyeglasses.
Face recognition systems work quite differently: find the eyes and measure geometry.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.