Memorable vs. Good Images

Yesterday’s post described an MIT algorithm that predicts a photo’s memorability at “near-human” levels. Today’s post shows preliminary findings using my own photos in a search for correlation, if any, between memorable & good images.

I’m curious about what correlation there might be between memorable images and images that a photographer or other viewers might like (memorable or not).

  • As a first step I ran 68 of my photos through the MIT algorithm.
  • These photos were from my on-line gallery where, from 2004 to 2009, I posted a photo-a-day (PAD)
    • They were from my gallery’s PAD Favorites collection
    • Each month the daily photo that got the most views was the month’s favorite
  • These 68 were representative of what both I, and viewers, thought were “good”
    • I asked the MIT algorithm “How memorable are they?”
    • I added the Memorability Score to each photo as shown below
    • Click on this next image to see the entire collection

memorable pbase


The first thing I learned from my PAD experiment is that landscapes are not memorable. You can see in the image above that two classics, photographed millions of times, are near the bottom of the memorability scale. This next photo of mine received Freeman Patterson’s (world acclaimed photographer) highest accolade. Another “good” but not too memorable photo.

memorable freeman


The most memorable landscape that emerged from my PAD favorite experiment just happened to be my photo that won an award for the

  • Advanced Color Print of the Year (from >1000+ entries over the course of a year)

harvest

This correlation between “good” (the PAD collection) and “memorable” (MIT) is what I am curious about. Could it be that:

  • Good image + Memorable image = Winning image (??)

Tomorrow I’ll show images with a perfect memorability score of 1.0.


Enter your email address to follow this blog and receive notifications of new posts by email.

2 thoughts on “Memorable vs. Good Images

  1. Happy New Year, Ed. If this MIT algorithm is able to distinguish a ‘memorable ‘ image, not necessarily a ‘good’ or ‘sings to me’ image, what is the future of photo competition? Not that I am a world class photographer, I stopped entering events due to lack of consistency from the same judges at different events. And this appears to be headed in the same direction. The lack of ‘memorability’ in an image is now reason for not being a winning image? How many ‘good’ images will go unrecognized because, if run through this algorithm, they are not ‘memorable’? While an interesting use of computer programming, I do not see this as a useful method to judge a photo. Hence, I will continue to make images that ‘sing’ to me and not be concerned that others can’t hear the tune.
    Very happy you are now settled in your new home, if one can ever be settled. Keep up with the thought provoking and informative articles.
    Ed Schoonover

    • I hear you – and agree. Hopefully the “memorable” half of the equation will never enter into judging (at least not explicitly; as for subconsciously, I share your opinion of many judges whose judgement makes my brain itch – which is why I stopped competing after 3 years).

      I’m not proposing that the algorithm should ever play a role in judging a competition. I’m merely curious as to whether some combination of Good + Memorable might help a competitor in screening candidates for entry. For example it would have correctly told me to enter the harvest scene and not the Freeman Patterson winner if I had to chose between them (although if I were the judge, I’d have given the FP the nod).

      Beauty is in the eyes of the beholder. This current series & interest of mine is just idle curiosity – and keeping my addled brain active.

      Happy New Year

Comments are closed.