Here’s 8 years ago I had a neural net try to “mimick” my face. After 10,000 tries, it failed.

 Here’s 8 years ago I had a neural net try to “mimick” my face. After 10,000 tries, it failed.
Yet you can get a sense of what it’s programmed to do.
Now 8 years later, they do much better. One of the things they introduced (I think Google did first) was using “Ground Truth” outlines in the training instead of the wobbly training like you see here.
So in this case, they’d probably have a generic “face”. Then an outline of mine. Then stretch the generic ground truth face to match mine. At that point, a perfect match of my face would be easier as it would just be colors and texture at that point which would be easy (it already could do that as you can see).
But composition? that part remains a mystery to me.
[responsivevoice_button voice="US English Male"]

Leave a comment

Your email address will not be published. Required fields are marked *


− 2 = two

Leave a Reply