Software and technology have done great things for society. There are robots that build cars, virtual-vision computers that let doctor perform microscopic remote surgery, self-driving vehicles that take the guesswork out of transportation, and more. But all of those advancements pale in comparison to the new AI-level software that helps you take the best selfie ever.
Using complicated parameters, Stanford University doctoral student Andrej Karpathy “taught” a computer in their Computer Vision Lab to evaluate selfies and judge their merits to determine if they were good or not. In almost exactly the same way that the human brain develops neural pathways for knowledge and decision making through repeated experiences, the computer “learned” how to judge a selfie by looking through more than 5 million of them and comparing the features of the selfie with the amount of positive feedback it received from viewers online.
According to a report by Mike Murphy for Quartz, the computer’s recommendations were interesting:
- We like women. The top 100 selfies the neural network chose were all women, most of which had long hair. The program also tended to prefer selfies that cropped out foreheads, for some reason.
- Selfies should be mostly face. All of the best selfies seem to show the person’s face taking up about one third of the photo, with the head tilted slightly.
- Distort the photo. Karpathy noticed that the vast majority of popular selfies had oversaturated the face in the photo, added some sort of filter, and probably added a border of some sort around the image.
Unfortunately, Murphy also reported that the computer did not think very highly of his selfie, and even went so far as to recommend a square image in the future. Anyone interested in submitting a selfie of their own for computer evaluation can Tweet it to @deepselfie and wait for a scored response. (Author’s Note: My own selfie was also dogged by the computer and received a score only 2.8 points higher than Murphy’s…sigh.)