Choosing the best photos has always been more a matter of art than of science, whether it's for a Facebook page or a marketing campaign. That may soon change, however, thanks to some new technology out of MIT.
Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have created an algorithm called MemNet that can predict how memorable or forgettable an image is almost as accurately as humans can, the university announced on Wednesday.
For each image, the algorithm produces a heat map showing which parts of the image are most memorable. The idea, ultimately, is to help increase images' memorability by emphasizing different regions.
The technology is now available through an online demo that anyone can try, and the researchers plan to turn it into an app as well, with a focus on helping users tweak their photos to make them more memorable.
Potential applications might include improving the content of ads and social media posts or developing more effective teaching resources, the researchers suggested.
“Understanding memorability can help us make systems to capture the most important information, or, conversely, to store information that humans will most likely forget,” said CSAIL graduate student Aditya Khosla, who was lead author on a related paper. “It’s like having an instant focus group that tells you how likely it is that someone will remember a visual message.”
Khosla is scheduled to present the paper in Chile this week at the International Conference on Computer Vision.
The research team behind MemNet had already developed a similar algorithm for facial memorability, but this new work uses techniques from deep learning, including neural networks to teach computers to sift through massive amounts of data to find patterns.
The team tested the algorithm on tens of thousands of images from several different data sets. Those images had already each received a “memorability score” based on the ability of human subjects to remember them in online experiments.
The algorithm then went head-to-head with humans by predicting how memorable a group of people would find a new, never-before-seen image. As it turned out, it performed 30 percent better than existing algorithms and was within a few percentage points of the performance of the average human.
Looking ahead, the research team next plans to try to update the system to be able to predict the memory of a specific person, and to better tailor it for individual industries such as retail clothing and logo design.
Google, Xerox and Nvidia are among the organizations that supported the research along with the National Science Foundation, the McGovern Institute Neurotechnology Program and the MIT Big Data Initiative at CSAIL.