220
SAME FUCKING FRAME! 4Chan Spec Ops went frame by frame and uncovered this GEM! (media.patriots.win)
posted ago by Pigpenlordofdirt ago by Pigpenlordofdirt +223 / -3
105 comments
105 comments share save hide report block hide child comments
Comments (105)
sorted by:
You're viewing a single comment thread. View all comments, or full comment thread.
▲ 2 ▼
– CertainlyNotEdward 2 points ago +3 / -1

Whoa holy shit, this is kinda nutso. I was on the "this is real" train before but I'm now leaning toward the "this is deep fake" train.

So a real person doesn't make nearly the same face twice, but DNNs do because their training dataset is finite and it doesn't learn in real-time to identify it already used a pose.

permalink save report block reply
▲ 1 ▼
– CertainlyNotEdward 1 point ago +1 / -0

(also NN wizards please don't pick me apart here; I know that's not how it works exactly, but my point I'm trying to get across is it's not like an LSTM: the network has no feedback to prevent over-reliance on a particular learned concept over time)

permalink parent save report block reply