Have a personal or library account? Click to login
AG-HybridNet: An Attention-Guided Hybrid CNN-Transformer Network for 3D Gaze Estimation Cover

AG-HybridNet: An Attention-Guided Hybrid CNN-Transformer Network for 3D Gaze Estimation

By: Yue Li and  Changyuan Wang  
Open Access
|Dec 2025

Figures & Tables

Figure 1.

The architecture of GazeTR
The architecture of GazeTR

Figure 2.

The detailed architecture of AG-HybridNet.
The detailed architecture of AG-HybridNet.

Figure 3.

The detailed architecture of the CNN branch.
The detailed architecture of the CNN branch.

Figure 4.

Schematic diagram of the Reparametrized Partial Convolution (RPConv) structure.
Schematic diagram of the Reparametrized Partial Convolution (RPConv) structure.

Figure 5.

Network architecture diagram showing the RPConv and TDConv Block structure.
Network architecture diagram showing the RPConv and TDConv Block structure.

Figure 6.

The detailed architecture of the attention mechanism.
The detailed architecture of the attention mechanism.

Figure 7.

MPIIFaceGaze and Gaze360 Loss Convergence Curves.
MPIIFaceGaze and Gaze360 Loss Convergence Curves.

Figure 8.

Comparison of Mean Angular Error on the MPIIFaceGaze Dataset.
Comparison of Mean Angular Error on the MPIIFaceGaze Dataset.

Figure 9.

Comparison of Mean Angular Error and FLOPs
Comparison of Mean Angular Error and FLOPs

Comparative Experimental Results on the Gaze360 Dataset

ModelMean Angular Error (°)
Full-Face14.99
Dilated-Net13.73
RT-Gene12.26
Gaze36011.40
Bot2L-Net11.53
Ours10.82

Comparative Experimental Results on the MPIIFaceGaze Datase

ModelMean Angular Error (°)
MPIIGaze5.40
Dilated-Net4.80
CA-Net4.10
AGE-Net4.09
GazeTR4.00
L2CS-Net3.92
Res-Swin-Ge3.75
Ours3.72

Comparison of Parameters and FLOPS for Different Models

ModelMean Angular Error (°)ParametersFLOPs
Dilated-Net4.803.9203.153
GazeTR4.0011.3941.834
Ours3.7221.2011.505
Language: English
Page range: 82 - 93
Published on: Dec 31, 2025
In partnership with: Paradigm Publishing Services
Publication frequency: 4 issues per year

© 2025 Yue Li, Changyuan Wang, published by Xi’an Technological University
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.