My reading of the paper is that it points out two things. First, current object detection methods do not confirm their detections by generating a matching image region (e.g., from a GAN). 1/
-
-
-
Second, the existing object detections employ a NON-LEARNED maximum suppression method to choose which of the many object detections to report. This is very unstable and causes detected objects to come and go as the elephant is moved around the image 2/
- 5 more replies
New conversation -
-
-
"can't" => "don't currently"
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
And what's your solution, Marcus? You've been preaching about the problems with deep learning for a while, as if the enlightened ones among us needed you to tell us that DL sucks. In fact, DL is completely irrelevant to AGI. But we're still waiting for your brilliant solution.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@LiadMudrik1 I wonder how the network will react to a chessboard in an oven (seriously!) -
Happy to share our stimuli and test! Would be very interesting indeed
End of conversation
New conversation -
-
-
The elephant should have been pink, for better comic effect.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Marcus doesn't know. This is not even the worst problem with DL. Unlike the brain, a DNN cannot see a pattern or object unless it has a prior representation of it in memory. This is a huge problem because it is crucial to survival. Neither Marcus not the DL gurus have a solution.
- 1 more reply
New conversation -
-
-
Anyone working in image recognition should experiment with his own brain, specifically serotonin analogues and serotonin re-uptake inhibitors. Layers of perception algorithm leek and overlay on top of picture. The brain does a pass of contrast wave and curve simplification.
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.