Appearance-Based Gaze Estimation Method Using Static Transformer Temporal Differential Network

Gaze behavior is important and non-invasive human–computer interaction information that plays an role in many fields—including skills transfer, psychology, interaction. Recently, improving the performance of appearance-based gaze estimation, using deep learning techniques, has attracted increasing attention: however, several key problems these deep-learning-based estimation methods remain. Firstly, feature fusion stage not fully considered: existing simply concatenate different obtained features into one feature, without considering their internal relationship. Secondly, dynamic can be difficult to learn, because unstable extraction process ambiguously defined features. In this study, we propose a novel method consider problems. We static transformer module (STM), which uses multi-head self-attention mechanism fuse fine-grained eye coarse-grained facial Additionally, innovative recurrent neural network (RNN) cell—that is, temporal differential (TDM)—which used extract integrated STM TDM with (STTDN). evaluated STTDN performance, two publicly available datasets (MPIIFaceGaze Eyediap), demonstrated effectiveness TDM. Our results show proposed outperformed state-of-the-art methods, including Eyediap (by 2.9%).