I love this bit of copy: "Neural network hallucinates missing details to make image look natural." The future is, in some ways, gloriously weird.
-
-
Show this thread
-
Here's a 2 second demo: my employee badge photo around the collar button, zoomed in at the same level. Left is the original; right is the LetsEnhance version at 4x zoom.pic.twitter.com/lGF2fPjQD7
Show this thread -
The computer, which does not know what a button is, has invented a plausible button that *does not actually exist in nature*, as a *freaking side effect.* (I don't have the actual shirt collar around to show you but *that button is not actually the button.*)
Show this thread
End of conversation
New conversation -
-
-
recover is wrong word
-
"credibly invent"
- 2 more replies
New conversation -
-
-
There's an app called TensorZoom for android that does something similar to this, it's based off google's TensorFlow and it's not half bad
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Neural nets may be cool, but their signup is definitely broken.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Someone else already mentioned it in your comments, but, this is *not* doing recovery. It is generating a plausible signal given its prior (obtained from the training set it was generated on). It's important to be clear on this since it has important implications.
-
For example, if someone tried to do facial reconstruction given a blurry image, you wouldn't get the original face, but you'd get *a* plausible face that is compatible with the pixels and the original training set.
End of conversation
New conversation -
-
-
I really don't think this should be optionalpic.twitter.com/jykM84Rkqb
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.