GazeFormer-MoE: Context-Aware Gaze Estimation via CLIP and MoE Transformer

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Zhao, Xinyuan, Chen, Xianrui, Chaddad, Ahmad
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917209244172288
author Zhao, Xinyuan
Chen, Xianrui
Chaddad, Ahmad
author_facet Zhao, Xinyuan
Chen, Xianrui
Chaddad, Ahmad
contents We present a semantics modulated, multi scale Transformer for 3D gaze estimation. Our model conditions CLIP global features with learnable prototype banks (illumination, head pose, background, direction), fuses these prototype-enriched global vectors with CLIP patch tokens and high-resolution CNN tokens in a unified attention space, and replaces several FFN blocks with routed/shared Mixture of Experts to increase conditional capacity. Evaluated on MPIIFaceGaze, EYEDIAP, Gaze360 and ETH-XGaze, our model achieves new state of the art angular errors of 2.49°, 3.22°, 10.16°, and 1.44°, demonstrating up to a 64% relative improvement over previously reported results. ablations attribute gains to prototype conditioning, cross scale fusion, MoE and hyperparameter. Our code is publicly available at https://github. com/AIPMLab/Gazeformer.
format Preprint
id arxiv_https___arxiv_org_abs_2601_12316
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle GazeFormer-MoE: Context-Aware Gaze Estimation via CLIP and MoE Transformer
Zhao, Xinyuan
Chen, Xianrui
Chaddad, Ahmad
Computer Vision and Pattern Recognition
Artificial Intelligence
We present a semantics modulated, multi scale Transformer for 3D gaze estimation. Our model conditions CLIP global features with learnable prototype banks (illumination, head pose, background, direction), fuses these prototype-enriched global vectors with CLIP patch tokens and high-resolution CNN tokens in a unified attention space, and replaces several FFN blocks with routed/shared Mixture of Experts to increase conditional capacity. Evaluated on MPIIFaceGaze, EYEDIAP, Gaze360 and ETH-XGaze, our model achieves new state of the art angular errors of 2.49°, 3.22°, 10.16°, and 1.44°, demonstrating up to a 64% relative improvement over previously reported results. ablations attribute gains to prototype conditioning, cross scale fusion, MoE and hyperparameter. Our code is publicly available at https://github. com/AIPMLab/Gazeformer.
title GazeFormer-MoE: Context-Aware Gaze Estimation via CLIP and MoE Transformer
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2601.12316