Saved in:
Bibliographic Details
Main Authors: Samaniego, Miren, Rodriguez, Igor, Lazkano, Elena
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.19936
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915511241015296
author Samaniego, Miren
Rodriguez, Igor
Lazkano, Elena
author_facet Samaniego, Miren
Rodriguez, Igor
Lazkano, Elena
contents We introduce CapStARE, a capsule-based spatio-temporal architecture for gaze estimation that integrates a ConvNeXt backbone, capsule formation with attention routing, and dual GRU decoders specialized for slow and rapid gaze dynamics. This modular design enables efficient part-whole reasoning and disentangled temporal modeling, achieving state-of-the-art performance on ETH-XGaze (3.36) and MPIIFaceGaze (2.65) while maintaining real-time inference (< 10 ms). The model also generalizes well to unconstrained conditions in Gaze360 (9.06) and human-robot interaction scenarios in RT-GENE (4.76), outperforming or matching existing methods with fewer parameters and greater interpretability. These results demonstrate that CapStARE offers a practical and robust solution for real-time gaze estimation in interactive systems. The related code and results for this article can be found on: https://github.com/toukapy/capsStare
format Preprint
id arxiv_https___arxiv_org_abs_2509_19936
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CapStARE: Capsule-based Spatiotemporal Architecture for Robust and Efficient Gaze Estimation
Samaniego, Miren
Rodriguez, Igor
Lazkano, Elena
Computer Vision and Pattern Recognition
We introduce CapStARE, a capsule-based spatio-temporal architecture for gaze estimation that integrates a ConvNeXt backbone, capsule formation with attention routing, and dual GRU decoders specialized for slow and rapid gaze dynamics. This modular design enables efficient part-whole reasoning and disentangled temporal modeling, achieving state-of-the-art performance on ETH-XGaze (3.36) and MPIIFaceGaze (2.65) while maintaining real-time inference (< 10 ms). The model also generalizes well to unconstrained conditions in Gaze360 (9.06) and human-robot interaction scenarios in RT-GENE (4.76), outperforming or matching existing methods with fewer parameters and greater interpretability. These results demonstrate that CapStARE offers a practical and robust solution for real-time gaze estimation in interactive systems. The related code and results for this article can be found on: https://github.com/toukapy/capsStare
title CapStARE: Capsule-based Spatiotemporal Architecture for Robust and Efficient Gaze Estimation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2509.19936