Google has developed a neural network NIMA, capable of evaluating images. Each person has his own preferences, so one can like a picture with the correct color rendition and white balance, and the other will close his eyes if the picture shows a beautiful flower or a cute animal. Algorithms of the search giant evaluate the overall quality of the image, based on the feedback of the users themselves.
“Our network can be used not only to evaluate images depending on the person’s perception, but also for various time-consuming subjective tasks such as intelligent photo editing, image quality optimization to attract users or minimize perceived visual errors in images,” says the Google blog .
The algorithm was trained in photographs, each of which was evaluated on average by two hundred people. After that, the network independently ranked the images, putting them an average score, like all respondents.
The neural network assigns images an estimate of 1 to 10, given certain factors. In other words, the same photo before and after processing can have a different rating. This technology will also find use in ranking photos by quality.
Google is confident that NIMA is suitable for working with other data, evaluating them in the same way as people.