PetaPixel

MIT Algorithm Tries to Predict How Many Likes Your Photo Will Get Per Day

MIT_3

A photographer’s primary concern when taking a photo might not be “I wonder how many likes this will get,” but being able to gauge popularity could still come in handy when you’re trying to decide which photos to upload to your favorite sharing site.

Enter MIT PhD candidate Aditya Khosla and his new algorithm that does just that: tells you how popular your photos will be before you even upload them.

Khosla works in MIT’s much-revered Computer Science and Artificial Intelligence Lab, and he recently used 2.3 million images from Flickr to create an algorithm that can rank the potential popularity of an image. Factors that make an image more popular include mini skirts, bright colors and people. Things that affect an image negatively: plungers and golf carts.

Screen Shot 2014-04-25 at 12.13.59 PM

These results are given to you on a scale from 0 to 10, which can then be converted to a potential “likes-per-day” on a log scale.

“If the score is 5, we expect roughly 2^5 = 32 views on your image per day,” Khosla tells The Verge. “If it’s 6, roughly 2^6 = 64.” You can even put a few of your own images through a stripped down version of the algorithm by clicking here.

I ran one of my own and Ben Von Wong’s photo that he took for Saving Eliza. Not surprisingly, the one by the professional photographer is expected to do better.

MIT_1

MIT_2

Of course, the stripped down version isn’t going to be nearly as accurate because it doesn’t take into account things like how many followers you have, how many groups you belong to and how long you’ve been a member (remember, this is based on Flickr) but it’s still interesting to see how a computer believes a photo’s content will affect its popularity.

To find out more, read the full paper by clicking here. And if you want to try it out for yourself, you can do so here. Fair warning though, the website has experienced a bit of traffic overload so wait times might be substantial.

(via The Verge via PopPhoto)


 
Get the hottest photo stories delivered to your inbox.
Get a daily digest of the latest headlines:
  • http://Taytus.com/ Roberto Inetti

    A photographer’s primary concern when taking a photo might not be “I wonder how many likes this will get,”

  • sdtransfertomich

    Well…maybe not a primary concern. But as a big user of Flickr, I do feel some validation when my photos make it to the Explore Gallery (top 500 downloads of the day). Photography is a self expressing art form, but if no one likes to look at the visual image you produce, then its only about yourself.

  • Frank McKenna

    Its an interesting concept. Unfortunately the scoring algorithm doesn’t appear to work on their site. This could have some great application out there.

  • http://www.vonwong.com/ Benjamin Von Wong

    such a weird system :O

  • http://www.pcmag.com Jim Fisher

    I think he needs some more research. A banal/awful family photo from a few years ago gets a 4.5, and a portrait I shot of 50 Cent during an interview got 4.4.

  • http://www.sin3rgy-creative.com/ David Liang

    I don’t know if it’s an error in the writing of the article or that of the creator of the algorithm but it doesn’t tell you if a “photo is going to be popular” it only tells you the likeliness that it will be based on passed history.

    What he’s basically done is created a ranking system based on parameters that have been popular in the past. Past performance is related but not indicative of future performance. It also doesn’t account for variables such as trends or fads. Neon t-shirts might not have been popular in the past but no one sees fads coming and they do happen.
    For an MIT student I’m sure he knows the Black Swan theory and should understand the limits of his algorithm.

  • http://www.sin3rgy-creative.com/ David Liang

    The writer of the article is misrepresenting the study for headline clicks. if you read the study they actually state areas of issue and don’t sound nearly as certain of their predictive ability as this writer makes it sound.

  • Ed

    “top 500 downloads of the day?….. You may wish to correct your arrogance

  • Eden Wong

    Sorry, but in my opinion this may be the biggest, “who the hell cares” post ever on this website.

  • nemomen

    Explore is the most “interesting” photos, though the interestingness algorithm doesn’t try to measure “good”, just “interesting.” It’s other people’s reactions to your photo that can be algorithmically measured (clicks, likes, comments), and which can also be gamed.

    I still like it when I get a photo Explored, though.

  • patricia666

    My Uncle Nathaniel recently got a nearly
    new red Chrysler 200 Sedan only from working part time off a home pc… find
    out this here C­a­s­h­F­i­g­.­ℂ­o­m

  • somebody

    Not bad.

  • sdtransfertomich

    I can see how one in 500 hardly sounds commendable, but given there are several million uploads a day……

  • sdtransfertomich

    Well, sure. With a photograph or any art, “good” is in the eyes of the beholder and not a measurable standard. Counting or measuring reactions is really the only way one can even try and measure if a photo gets noticed.

  • sdtransfertomich

    OK. That’s nice. Did your uncle get a “like” from the photo he posted of his car?