Photography by the numbers: MIT algorithm analyzes photos, predicts their popularity
posted Wednesday, May 21, 2014 at 5:35 PM EDT
Ever wondered what it is that makes some of your photos popular? You're not alone: MIT student Aditya Khosla wondered the same thing, as did his fellow researchers, Atish Das Sarma of eBay, and Raffay Hamid of DigitalGlobe. Where you likely differ from these three gentlemen, though, is that you probably only wondered about it, whereas they promptly set to work on an attempt to find the answer. And not just that -- they also wanted to see if they could predict the popularity of photos before they were proven as viral hits or duds in the real world.
In their recently-published 10-page paper, "What makes an image popular?", Kohsla, Das Sarma and Hamid outline their findings and the theory behind them, as well as the photo popularity tool their research led them to create. In a nutshell, the tool works by comparing uploaded photos to a corpus of 2.3 million images culled from Flickr, looking for similar images and considering how many views they've had. The paper proposes going a step further by looking at social cues for the uploader, such as how many friends they have, an acknowledgement of the fact that your photos' popularity depends to a significant extent on your own popularity.
But does it do a good job? We'll leave that for you to decide. We uploaded a variety of photos to try the service out, ranging from intentionally-bland cellphone images to some more interesting shots, with varying degrees of editing. In the captions are the results we got for each image. Do you think the tool gets its ratings right? Sound off in the comments below...