The direction of human gaze is an important indicator of human behavior, reflecting the level of attention and cognitive state towards various visual stimuli in the environment. Convolutional neural networks have achieved good performance in gaze estimation tasks, but their global modeling capability is limited, making it difficult to further improve prediction performance. In recent years, transformer models have been introduced for gaze estimation and have achieved state-of-the-art performance. However, their slicing-and-mapping mechanism for processing local image patches can compromise local spatial information. Moreover, the single down-sampling rate and fixed-size tokens are not suitable for multiscale feature learning in gaze estimation tasks. To overcome these limitations, this study introduces a Swin Transformer for gaze estimation and designs two network architectures: a pure Swin Transformer gaze estimation model (SwinT-GE) and a hybrid gaze estimation model that combines convolutional structures with SwinT-GE (Res-Swin-GE). SwinT-GE uses the tiny version of the Swin Transformer for gaze estimation. Res-Swin-GE replaces the slicing-and-mapping mechanism of SwinT-GE with convolutional structures. Experimental results demonstrate that Res-Swin-GE significantly outperforms SwinT-GE, exhibiting strong competitiveness on the MpiiFaceGaze dataset and achieving a 7.5% performance improvement over existing state-of-the-art methods on the Eyediap dataset.

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