This is bringing our staff back to the days of when high school lists would float around the school, telling us which kids were “hot or not.”

Instead of a list, we now have a selfie bot that will rate your selfie.

Andrej Karpathy is part of a project involving deep neural networks. On his blog he writes, “Convolutional Neural Networks are great: they recognize things, places and people in your personal photos, signs, people and lights in self driving cars, crops, forests and traffic in aerial imagery, various anomalies in medical images, and all kinds of other useful things.”

And because Convolutional Neural Networks are that awesome, he did an experiment that took a 140-million-parameter state-of-the-art CNN and fed it 2 million selfies, training it to see what a good selfie was and what a bad selfie was.

Just maybe we can all learn how to take a selfie. We tried it out here at SD Times, and here are our selfies (with scores).

1030.sdt-blog-mandyOnline and Social Media Editor Madison tried for a filter and a tilted-head shot. Selfie Bot scored her with a 49.5%, a little below average.

1030.sdt-blog-selfie-christinaOnline and Social Media Editor Christina tried a “harrumph” face. Selfie Bot rated her with a 46.1% and said, “Hey, it could be worse.”


Winner for best selfie is Copy Editor Adam LoBelia, who got a 52.5% (even though it was taken under duress). “Not bad!”

Karpathy found a few patterns that stood out when looking at the collection of the top 1000 selfies. Some include: being female, cutting off your forehead (just in the picture, not real life), oversaturating the face, and putting a filter on it.

Interested in what your own selfie rates? Visit Karpathy’s blog for more information, or tweet to @deepselfie with a link or a picture of your selfie, and it will rate you within a minute.