Computer Vision News - February 2022

9 Auditing Saliency Cropping Algorithms to see which element the algorithm chose to present when cropped. It favoured McConnell almost every time. “ Many people don’t realise these kinds of algorithms are created and deployed almost everywhere in these big social media and technology companies , ” Abebawarns. “ Evenwithin visionresearch , people don’t know about it, because it’s kept quiet, and the code and data aren’t open access. It’s difficult to carry out this kind of work because most of the datasets are protected by proprietary rights. ” The team carried out two experiments. The first was on the male-gaze-like artifacts. “ We created a Twitter account and put images through the platform to see what came out, ” Abeba explains. “ We kept noticing when images of women on the red carpet at the Emmy Awards, for Ethiopian-born Abeba is a cognitive scientist by training but says she has been drawn more and more to the computer vision side of AI. Her associates on this work, Vinay Uday Prabhu and John Whaley from UnifyID Labs, are two experts in computer vision. “ My work sits at the AI end of cognitive science, ” she tells us. “ You can find all kinds of fascinating intersections between computer vision and cognitive science. Within this research, I’m particularly interested in how vision models present intelligence, emotion, or interestingness. ” The first the team knew that platforms such as Twitter use saliency cropping algorithms was when an image of US Senator Mitch McConnell and former President Obama went viral for all the wrong reasons. A Twitter user put photos of the pair at either end of a blank image A collage of real-world user-uploaded images on Twitter that exhibited male-gaze-like artifacts