A new algorithm was created in order to predict an image's memorability rate, which means it could determine whether your photos will be a hit or be easily forgotten.
MIT's team of researchers at the Computer Science and Artificial Intelligence Lab (CSAIL) claimed that the algorithm performs 30 percent better than current algorithms and has only a few percentage points difference from the average performance of a human.
"One hallmark of human cognition is our massive capacity for remembering lots of different images, many in great detail, and after only a single view," said the researchers in the introduction part of the paper. "Images that are consistently forgotten seem to lack distinctiveness and a fine-grained representation in human memory. These results suggest that memorable and forgettable images have different intrinsic visual features, making some information easier to remember than others."
To further encourage more research on what the team described as an under-studied topic in computer vision, they have also published the so-called LaMem database which now contains 60,000 images. Each of the images has been annotated and has included detailed metadata on specific qualities that include emotional impact and popularity.
"The basic idea of our novel procedure is to allow the second occurrence of an image to occur at variable time intervals," wrote the researchers. "This procedure is based on the finding that the memorability ranks of images are time-independent. Using this new experimental setting, we build a novel massive memorability dataset, with scores on 60,000 images while keeping a low cost."
When asked about the type of patterns that the MemNet algorithm tries to identify when it wants to predict the image's memorability or forgettability, Aditya Khosla, PhD candidate at MIT CSAIL and lead author of the paper, explains:
"This is a very difficult question and active area of research. While the deep learning algorithms are extremely powerful and are able to identify patterns in images that make them more or less memorable, it is rather challenging to look under the hood to identify the precise characteristics the algorithm is identifying."
Khosla added that based on an initial analysis, images that tend to be highly memorable are exposed human body parts and faces. Likewise, other images such as beaches, the horizon or any outdoor scenes tend to be forgettable.
The team plans to launch an app in the future, which will allow users how to enhance their images in order to increase their impact. For now, interested customers can try a demo. They also mentioned certain "tricks" that users can apply based on algorithmic results. These include using a heat map that would blur out non-memorable regions of an image in order to emphasize those that have high memorability, applying a filter similar to how it's done through Instagram, cropping the image and adding or removing certain objects from the image.
