YourPose-CNN: A Multi-Head Mask-Guided Unified Framework for Head Pose Estimation Using ZetaLite U-Net

The research work indicated here involves the development of unified framework which is capable of performing both discrete as well as continuous head pose estimation.In the first stage the face is detected through Sample and Computation Redistribution for Face Detection (SCRFD).In the second stage the Nose-Lip region mask is constructed for detected face through ZetaLite U-net model, a memory efficient and powerful U-net designed for efficient Nose-Lip mask extraction.In the third stage YourPose-CNN(a multi task multi head CNN) is designed which concatenates the detected face with the nose-lip mask for estimating the Euler angles as well head pose class among seven classes considered in this research study.On the BIWI dataset, ZetaLite U-Net attained higher training and validation accuracy than the conventional U-Net (99.01%vs. 96.40%and 98.29% vs. 93.80%)and a comparable validation IoU (87.48% vs 87.56%), while using far fewer parameters.The proposed model achieves an overall MAE of 4.0671° for regression and a classification accuracy of 94.84% on the BIWI test dataset, where the training validation and testing partitions are completely subject disjoint.Its potential for use in a variety of video analytics applications is further demonstrated by the cross-domain evaluation carried out on various image and video-based datasets, as detailed in this study.

Authors

Institutions

Publication Details

Journal
International journal of intelligent engineering and systems
Published
2026-11-17
DOI
https://doi.org/10.22266/ijies2026.1231.54
Primary Topic
Face recognition and analysis
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

YourPose-CNN: A Multi-Head Mask-Guided Unified Framework for Head Pose Estimation Using ZetaLite U-Net

Shet Reshma Prakash, V N Manju
International journal of intelligent engineering and systems
Face recognition and analysis
article

YourPose-CNN: A Multi-Head Mask-Guided Unified Framework for Head Pose Estimation Using ZetaLite U-Net

Shet Reshma Prakash, V N Manju
article en

Abstract

The research work indicated here involves the development of unified framework which is capable of performing both discrete as well as continuous head pose estimation.In the first stage the face is detected through Sample and Computation Redistribution for Face Detection (SCRFD).In the second stage the Nose-Lip region mask is constructed for detected face through ZetaLite U-net model, a memory efficient and powerful U-net designed for efficient Nose-Lip mask extraction.In the third stage YourPose-CNN(a multi task multi head CNN) is designed which concatenates the detected face with the nose-lip mask for estimating the Euler angles as well head pose class among seven classes considered in this research study.On the BIWI dataset, ZetaLite U-Net attained higher training and validation accuracy than the conventional U-Net (99.01%vs. 96.40%and 98.29% vs. 93.80%)and a comparable validation IoU (87.48% vs 87.56%), while using far fewer parameters.The proposed model achieves an overall MAE of 4.0671° for regression and a classification accuracy of 94.84% on the BIWI test dataset, where the training validation and testing partitions are completely subject disjoint.Its potential for use in a variety of video analytics applications is further demonstrated by the cross-domain evaluation carried out on various image and video-based datasets, as detailed in this study.

International journal of intelligent engineering and systemsVol. 19(12)
Visvesvaraya Technological University (IN)
Face recognition and analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon