Twitter has long had the feature where users can upload photos onto its platform, but the company also has a problem because when photos are uploaded, they are cropped rather oddly in their preview. This means that sometimes at a glance it might be hard to tell what’s going on in the photo.However this is something that Twitter wants to fix and are using AI to help make image crops “smarter”, where the previews of images will focus on an aspect of the photo that it thinks will interest users the most, which is also known as “saliency”.
According to Twitter, “In general, people tend to pay more attention to faces, text, animals, but also other objects and regions of high contrast. This data can be used to train neural networks and other algorithms to predict what people might want to look at. The basic idea is to use these predictions to center a crop around the most interesting region.”
Twitter notes that as it stands, neural networks that are used to predict saliency can be slow, which is why they have used knowledge distillation that will train a smaller neural network to imitate a larger and more powerful one. Basically the brunt of the work will be done by the larger network, and the results will then be passed onto the smaller network.
These changes are currently in the process of being rolled out to Twitter’s users and will be applicable across the web version of Twitter, as well as its iOS and Android counterparts.
Filed in . Read more about AI (Artificial Intelligence) and Twitter.