I particularly like how it can be composed with other algorithms (face detection has been mentioned) to change the energy function and make it even smarter. It's one of those things you can implement in a weekend and have a lot of fun with.
Kudos to everyone involved in developing this!
This is actually the case for many such discoveries - once you understand them, and furthermore realize the simplicity behind the result, you wonder why it wasn't discovered earlier.
In some cases, it actually was - sometimes many times over by different people. The only reason it didn't catch on is likely due to societal factors, technology (for instance, many algorithms that were later used to great effect for certain work today were actually known much, much earlier - sometimes almost to the dawn of electronic computing, or further back - but the technology to exploit them didn't exist; they were mostly "paper only", and sometimes only in a simplified form suitable for hand calculation, or only done for so many few iterations), or lack of application. Computed tomography is - I think - an example (among many) in this space.
Some discoveries and such sit around for a long time until society, technology, or application need "catches up"; during that time they may be "forgotten". Other times, they are published - but in the wrong place, or time, or whatever - and never reach the critical amount of audience to perpetuate them further.
One example of this that I know of happened with a graphics discovery on one of my childhood 8-bit computers; that discovery, had it been known more widely, might have made a greater impact within the community at that time. But while it was known then, it was at the wrong time and was published in the wrong magazine. Ultimately what doomed it was likely the fact that the transition from "home computers" to "PC revolution" occurred. So the discovery had to wait - and it was "rediscovered" by retro enthusiasts of the platform (many of which were now "old hands" of the same machine - who had been publishers, business owners, programmers, etc - in the machine's heyday), about a decade ago. So it had to wait about 20 years or so.
Some of these discoveries, though, really make you wonder, in a "what if" kind of manner.
The steam engine is one of them; it is such a simple device that a rudimentary form could have been built by the ancient Greeks; all the basic technology was there, at least for a low-pressure system.
But all that was known and created was arguably a toy, a curiosity, the aeolipile (Hero's engine). Even it was based on knowledge that had languished for about 200+ years. For many reasons, it was never developed further, nor was anything involving cylinders or whatnot created (although Hero certainly knew and understood enough to take things further).
What simple things are we currently sitting on - staring us in the face - but yet we do nothing with them, but if we did, might propel our society and culture forward (by anything of decades to potentially much more)? Can we identify them? Why do we not exploit them if we can?
In a way, it's analogous to so-called "black swans" - that seem obvious in retrospective, but are likely impossible to identify prior to that point. These things don't keep me up at night, but they are interesting to ponder.
I'd be interested in learning more about that, if I don't already know about it. Was it something to do with the C64?
$ apt install gimp-plugin-registry
I especially liked the linked demonstration video at: https://www.youtube.com/watch?v=vIFCV2spKtg .
The problem is, the dynamic programming and/or median extraction techniques that the algorithm uses (IIRC) are not very differentiable.
My question here is: once #4 has been completed how do you 'merge' the pixels that are left once the seam pixels have been 'discarded'?