I love math. So do the folks at MIT. They've created a remarkable algorithm that can analyze a photo, assigning it a memorability score and analyzing exactly which parts of it are most memorable. And now, you can try it for yourself.
Computational photography is a very big part of the future of photography. Whether it's using software to correct lens aberrations or creating crazy 16-lens cameras, the increasing power of computers working in tandem with camera hardware has opened new possibilities that will revolutionize many areas of photography in years to come. Now, researchers from MIT's Computer Science and Artificial Intelligence Laboratory have created an algorithm that can predict a photo's memorability with "near-human" accuracy.
The "MemNet" algorithm holds potential in many areas of application: better advertising, social media, filmmaking, and improving human memory. The algorithm uses a "deep-learning" approach, one of the most exciting areas of artificial intelligence, in which a computer uses "neural networks" to learn to infer patterns without additional input from humans. The researchers hope to develop the algorithm into an app that can subtly improve images, but have also noted applications in improving human memory. I tested it with a few of my own images and was wholly impressed.
Memorability Score: .785
Memorability Score: .637
Care to try out MemNet for yourself? Check it out here!
[via MIT News]
ehh..
So this could be interesting if it worked. And I was hopeful. But in these examples, it's seems like they're just putting a heat map around identified faces. Usually, that would simply be the memorable part. But here are example images with no background anyway. And when something like a multiple-exposure cello bow is obviously more memorable (as the defining portion of this photograph, one might say) than her almost expressionless face, I really have to wonder when I see the "heat map" of what's more memorable... Hmm. And if I have to write "memorable" one more time, going to die. :-)
I must be a photography God! I scored .947 on one of my headshots! Lowest score was a far away shot of a couple in a field .241. It seems to like close up shots that have a clean background.
Hmm, I wonder if Flickr will buy/license this. They have a patent on photo "interestingness" which is what determines who's photo goes into their coveted Explore.
So what made you go pro and shoot weddings for $5 an hr,
Well i consistantly score above 90% on the MIT app,
and my mom said i was good on facebook.
Lets see when the first about page features this, where someone uses this along side the bogus "award winning" tag :)
Well...i'm really good, lowest score .748, highest .956...am i that good??
What a load of b0llocks... MIT should concentrate more on how to create ANTMAN.
This pretty much sums it up....
Well said.
Sadly it doesn't work that reliable. A picture of a street sign scores 85% and a picture from a roadtrip to scotland where we climbed a mountain and took a picture on top of it of the three of us just scored 32%.
Ok it's about memorability of a picture not the memory behind it. But I will not remember a street sign more than a vast landscape with me and my friends in it.
This is an interesting start. It does seem to focus mostly on faces and it finds color images more memorable than black and white. I have run several Ansel Adams photos through and they all come out with low or very low memorability scores.
Crap
Science you failed hard this time.