(bounding box), directly in the browser. Underneath, it uses a trained convolutional neural network that provides bounding box predictions for the location of hands in an image. The convolutional neural network (ssdlite, mobilenetv2) is trained using the tensorflow object detection api