How can any photographer take an image that will never be forgotten? For starters, quit taking nature photos, whether it’s a sunset, mountain range, lake or ocean. Those images are among the least distinctive, according to an MIT algorithm trained to recognize memorable photographs nearly as well as humans.
Amateur photographers with dreams of taking memorable images shouldn’t trek to national parks and obsess over the best composition and shutter speed. Staying at home with a goofy friend — or venturing to a quirky furniture store would deliver more powerful results.
What’s really memorable? Weird and counterintuitive images do well, according to the algorithm. A stop sign spray-painted with the word “go,” a peculiar chair, a man in a sumo suit dragging a snowboard or a shirtless man wearing a bear mask.
Conflict and suspense win — be it a pair of scissors in flight — or a man standing in front of a bear. Photos of bedrooms generally induce yawns.
You can upload your own photos on the researchers’ website to see how they score.
One of my favorite photos I took recently — a beautiful image of a sunrise — was ranked as “very low” on the memorability scale. The algorithm deemed that the most interesting thing about the image was actually a row of homes under construction. A throwaway image of mine — Silly Putty made to resemble the poop emoji — scored “very high.”
To gauge the memorability of photos, people were shown hundreds of photos and asked which images were shown to them a second time. Images that were recalled on their second appearance were considered memorable.
The researchers took the results from the human tests and used it to train a computer system. The algorithm then began to realize the underlying traits that made the image memorable to humans.
The precise characteristics that the system is identifying are not always known. Such is the nature of deep learning — a hot area in tech in which computers are trained to recognize patterns. It’s the backbone of breakthroughs such as computers that can teach themselves to paint like Pablo Picasso, and cars that can drive like humans.
Aditya Khosla, the lead author of the research, thinks his findings could have positive implications for education. The team found that peoples’ memory improves when they are shown a higher percentage of memorable images. Textbooks and teaching aids could start to use visual aids that have been proven to stick in our heads.
There appears to be one big exception to the algorithm’s powers — iconic historical photos.
Take for example “Tank Man,” the iconic image of a man standing in front of a line of tanks during the 1989 Tiananmen Square shootings. The algorithm’s heat maps suggests that the light fixture in the foreground is the most interesting aspect. And overall, the images’s memorability ranks “low.”
Tiananmen Square isn’t the only historical image to score surprisingly low. A photo of the moon landing only gets a “medium memorability” rating, as does the flag raising on Iwo Jima. The famous portrait of George Washington crossing the Delaware River scores even worse.
Khosla says his team’s algorithm may struggle to grasp the context of such images and the emotions they evoke.
“A space landing, that is something historic. It has too much back meaning attached to it,” Khosla said.