Sign In

Communications of the ACM

ACM TechNews

This Is the Best Way to Take a Selfie, According to Artificial Intelligence

Taking a selfie.

A doctoral student at Stanford University has trained a neural network to judge selfies.

Credit: kaisorn/

Stanford University Ph.D. student Andrej Karpathy has trained a neural network to determine what constitutes a good selfie.

Karpathy, a researcher in the Computer Vision Lab, fed the system 2 million selfies from around the Web to build up its knowledge. The program analyzed the images, breaking them down into layers of shapes and colors, and learned to evaluate the quality of a selfie based on the number of likes they had.

Karpathy then showed the neural network 50,000 photos it had never seen before to test its new knowledge base. He says the program produced some interesting insights on what humans like in terms of selfies.

The most popular selfies are of women, show mostly a person's face, and have a filter, such as a border around the image. The neural network also determined people tend not to like selfies taken in low light, taken too close to the camera, and group shots.

People now can tweet their selfies to Karpathy's Twitter bot for evaluation.

View Full Article


Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account