Project information
Many human-computer or human-robot interactions require the capability of the system of
understanting whether the user is paying attention or not. However, to train such systems,
large amounts of data are needed, but they are currently unavailable. In this paper, we
first address the issue of data scarcity by creating a large dataset -- with about 120k
images -- for the attention detection task. Then, we develop a strong baseline system which
is able to correctly perform the task, achieving competitive results on the proposed
dataset. Further, we extend our system by: i) adding an auxiliary face detection module, and
ii) introducing a novel GAN-based data augmentation technique. Finally, we design a web
application to enable real-time testing of the developed model.
