Face authentication, in the context of privacy for phones, has been explored for some time. However, face recognition alone is not enough when you want to have private online conversations or watch a confidential video in a crowded space where there are many other people present. Each of them may or may not be looking at your private content displayed on your device, e.g. a smart phone. Because of the quick, robust, and accurate gaze detection mobile model we can now easily identify the face identity and gaze simultaneously in real time. Hence, the application, an electronic screen protector, can enable its enrolled users to continue reading private and confidential contents on your mobile device, while protecting their privacy from onlookers in a crowded space such as the subway or an elevator.
We enable this by transfer learning from one mobile model to a different, but related task. Our final multihead mobile model is robust under varying lighting conditions and head poses. The runtime is 2ms per face for gaze detection (DetectGazeNet), 47ms per face for face recognition, and 115ms per frame for face detection in average.