Back in 2018 I wanted to turn this into a more ambitious article but that never happened. I do like the plots, though, so I’m publishing it mainly to show them off.

---

At some point in 2012 or 2013, something weird happened. After years of nothing, object detection accuracy started going up.

Then it brought with it speech recognition.

And image recognition.

What happened in 2012? This.

Green dots mean deep learning.

It’s fascinating to see that in 2012, only one or two attempts use deep learning, but they are so far above the ceiling implied by the traditional techniques, that the next year almost everyone uses deep learning.

Rarely do new techniques conquer so quickly (over just a year!) but I’m kind of happy to see that when they deserve it, they do.