The goal was to determine if an algorithm could guess the popularity of a photo by analyzing what is in the shot and by looking at the network of the photographer.
It turns out predicting if a photo is going to be popular by looking at its objective aspects like color, texture, gradient, and even object-recognition, is still in its infancy. When they mixed 100 photographers and their 1.4M shots, their algorithm was right a little bit more than one-third of the time. This means asking a blind person to do the same would yield better results.
Everything changes when social cues are added to the set. The number of views can predict if an image will be liked 62% of the time, while the tags by themselves help guessing right for 40% of photos.
Altogether, social and contextual cues can predict the popularity of a photo 2 times out of 3.
If we add image features with social context, that number goes up to 72%.
For photographers, this means there is a strong correlation between image popularity and your network. For a photographer without any network, it is harder to know if a photo will be popular just by looking at the objects in the photo, the colors or other technical aspects.
That being said, a simple rule is right most of the time: photos with people are in general more liked than sceneries or objects.
The good news is, if your images are not liked as much as you’d want to, it might have nothing to do with your talent as a photographer. You need to grow your viewership. Getting more visibility might be the key to making your images more popular.
You can try the fun tool they researchers made public, and don’t forget to take the results with a grain of salt.
Photo by Nicola since 1972
This post was orginally published on Crowdmedia’s blog
Author: Martin Roldan
I believe democratization of media through images and video is going to change the way we consume information. I co-founded CrowdMedia to bring this vision to reality. Winter is for snowboarding, summer for Wakesurfing and BBQ. And the whole year is for family.