How To Take The Perfect Selfie, According To Science

Let us get over the fact that we care about taking the perfect selfie and get down to business, shall we?

Andrej Karpathy, a computer science graduate student at Stanford University, tasked an image-recognizing deep neural network to determine what it is that makes a good selfie. His findings? Women who abide by the rule of thirds, use a filter and let their long locks fall over their shoulders achieve the greatest selfie success.

1.2 million likes: Not too shabby.

Karpathy trained a Convolutional Neural Network, a type of data mining network capable of processing 140 million different specifications, to judge whether a selfie was successful or not. He began his experiment by running a script through ConvNet to collect images tagged with #selfie. He then whittled his database of more than 5 million photos down to 2 million, all of which contained at least one face.

Next, Karpathy analyzed the number of likes per followers, labeling the ones with the highest rates as good selfies, and giving those with the least number of proportional “likes” a negative label. “I took all the users and sorted them by their number of followers,” Karparthy wrote on his blog. “I gave a small bonus for each additional tag on the image, assuming that extra tags bring more eyes. Then I marched down this sorted list in groups of 100, and sorted those 100 selfies based on their number of likes. I only used selfies that were online for more than a month to ensure a near-stable like count.”

Once it was clear his ConvNet parameters were working, Karpathy ran a batch of 50,000 previously unanalyzed selfies —> Read More