MIT created a tool that will tell you how memorable your photos are

Courtney Carmody

Every day, we take countless photos of our food, places we visit, our friends, and, yes, ourselves. They're supposed to help us remember our lives. However, most of these photos tend to be wiped from our memory soon after we snap them.

The reason is that not all photos are created equal. Some are more forgettable than others. Thanks to a handy new tool created at MIT, you can test how memorable they are. You upload a photo, and it spits out a memorability score, what the researchers claim is the likelihood that you'll actually commit that image to memory. It's a fun tool. You can test it out for yourself.


I decided to try it out on the some of the Republican Presidential candidates, in honor of Tuesday's debate:

Donald Trump
Ben Carson
Ted Cruz

and three of my favorite U.S. cities: New York, San Francisco and Miami:


This isn't very surprising. After all, we're wired to remember faces, a point this tool bears out.

"People are more memorable. Body parts of semi-clothed people are more memorable. Not surprising, right? The things that are very forgettable are open scenes [like landscapes]," Aditya Khosla, one of the MIT researchers who worked on the project, told me. "That's probably why we use them as wallpaper—so we don't clutter our brains with useless information."


The computer, the researchers say, is able to predict whether people are likely to remember or forget an image, and that it can do it to near-human levels.

"We're not saying that we can predict what you'll remember [as an individual], but that what people will remember on average is going to be very similar," he said.


The app is based on research done at MIT's Computer Science and Artificial Intelligence Laboratory to try to figure out what makes images sticky, on average. Scientists showed people, on the online jobs platform Mechanical Turk, a series of images, some of which were repeats, and asked them to press a button when they thought they saw a picture they'd already seen. Based on that, they assigned a photo a memorability score. If 70 people out of a hundred remembered it, for example, it's score was 0.7. The higher the score, the more memorable it was. They fed this data, along with 60,000 images, into a computer so it could learn what kinds of photos people tended to remember and which they forgot. Then, they tasked it with predicting the memorability score of an image it hadn't seen before. That's what I did with the images of the Republican candidates and American cities.

The app, which is called LaMem, also creates a heatmap that homes in on the most memorable parts of an image, as you can see above. (Red signifies most memorable) Again, as you might expect, the faces in the images of the Republican candidates are the most impactful parts.


This is a potentially important piece of information to have. If advertisers, designers or educators who want to create products or content that stick in people's mind, they could use a tool like this to tweak their designs to make them more memorable. It could also be useful in helping create popular memes, the visual currency of the internet.

For reasons we don't fully understand, some faces are also more memorable than others, Khosla says, so LaMem could be used to aid casting directors in the film and fashion industries find new talent.


Right now, it's not quite up to that task though. "It's good for images in general, but not for specific categories," he told me. If you gave it 100 faces or potential logos for a company and asked LaMem to rank it, it would spit back scores, but they wouldn't be as reliable. (Sorry, Carson, you not actually the most memorable.) "We would need to build databases of that sort to make that kind of comparison much more reliable."

If and when it actually does that, it'll be another instance of technology changing how we perceive the world and remember it.


Daniela Hernandez is a senior writer at Fusion. She likes science, robots, pugs, and coffee.

Share This Story