This project is based on the Labeled Faces in the Wild Attributes+ dataset released with the 2015 paper Deep Learning Face Attributes in the Wild by Ziwei Liu et al. covering 73 face attributes. "Big Lips", "Bushy Eyebrows", "Double Chin", four racial groups, "Sunglasses", "Curly Hair"... We trained a MobileNetV2 network on LFWA+ categories and then used the embeddings to add a few more categories using custom datasets. This all comes together in 10 minute game that explores what it feels like to be described by the machine, and placed in some of these weird categories. All the analysis runs in-browser with TensorFlow.js on mobile and desktop, no images are sent to any server. Usually this analysis happens behind the scenes, but we wanted to create an experience where people had the chance to see it play out in realtime and build direct intuition.