From One Domain to Another: The Pitfalls of Gender Recognition in Unseen Environments

Gender recognition from pedestrian imagery is acknowledged by many as a quasi-solved problem, yet most existing approaches evaluate performance in a within-domain setting, i.e., when the test and training data, though disjoint, closely resemble each other. This work provides the first exhaustive cro...

Full description

Saved in:
Bibliographic Details
Main Authors: Nzakiese Mbongo, Kailash A. Hambarde, Hugo Proença
Format: Article
Language:English
Published: MDPI AG 2025-07-01
Series:Sensors
Subjects:
Online Access:https://www.mdpi.com/1424-8220/25/13/4161
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849704069102305280
author Nzakiese Mbongo
Kailash A. Hambarde
Hugo Proença
author_facet Nzakiese Mbongo
Kailash A. Hambarde
Hugo Proença
author_sort Nzakiese Mbongo
collection DOAJ
description Gender recognition from pedestrian imagery is acknowledged by many as a quasi-solved problem, yet most existing approaches evaluate performance in a within-domain setting, i.e., when the test and training data, though disjoint, closely resemble each other. This work provides the first exhaustive cross-domain assessment of six architectures considered to represent the state of the art: ALM, VAC, Rethinking, LML, YinYang-Net, and MAMBA, across three widely known benchmarks: <span style="font-variant: small-caps;">PA-100K</span>, <span style="font-variant: small-caps;">PETA</span>, and <span style="font-variant: small-caps;">RAP</span>. All train/test combinations between datasets were evaluated, yielding 54 comparable experiments. The results revealed a performance split: median in-domain F1 approached 90% in most models, while the average drop under domain shift was up to 16.4 percentage points, with the most recent approaches degrading the most. The adaptive-masking ALM achieved an F1 above 80% in most transfer scenarios, particularly those involving high-resolution or pose-stable domains, highlighting the importance of strong inductive biases over architectural novelty alone. Further, to characterize robustness quantitatively, we introduced the <i>Unified Robustness Metric</i> (URM), which integrates the average cross-domain degradation performance into a single score. A qualitative saliency analysis also corroborated the numerical findings by exposing over-confidence and contextual bias in misclassifications. Overall, this study suggests that challenges in gender recognition are much more evident in cross-domain settings than under the commonly reported within-domain context. Finally, we formalize an open evaluation protocol that can serve as a baseline for future works of this kind.
format Article
id doaj-art-a4ed1558deba4f908ba2f20640e5e4e7
institution DOAJ
issn 1424-8220
language English
publishDate 2025-07-01
publisher MDPI AG
record_format Article
series Sensors
spelling doaj-art-a4ed1558deba4f908ba2f20640e5e4e72025-08-20T03:16:56ZengMDPI AGSensors1424-82202025-07-012513416110.3390/s25134161From One Domain to Another: The Pitfalls of Gender Recognition in Unseen EnvironmentsNzakiese Mbongo0Kailash A. Hambarde1Hugo Proença2Department of Computer Science, University of Beira Interior, 6201-001 Covilhã, PortugalDepartment of Computer Science, University of Beira Interior, 6201-001 Covilhã, PortugalDepartment of Computer Science, University of Beira Interior, 6201-001 Covilhã, PortugalGender recognition from pedestrian imagery is acknowledged by many as a quasi-solved problem, yet most existing approaches evaluate performance in a within-domain setting, i.e., when the test and training data, though disjoint, closely resemble each other. This work provides the first exhaustive cross-domain assessment of six architectures considered to represent the state of the art: ALM, VAC, Rethinking, LML, YinYang-Net, and MAMBA, across three widely known benchmarks: <span style="font-variant: small-caps;">PA-100K</span>, <span style="font-variant: small-caps;">PETA</span>, and <span style="font-variant: small-caps;">RAP</span>. All train/test combinations between datasets were evaluated, yielding 54 comparable experiments. The results revealed a performance split: median in-domain F1 approached 90% in most models, while the average drop under domain shift was up to 16.4 percentage points, with the most recent approaches degrading the most. The adaptive-masking ALM achieved an F1 above 80% in most transfer scenarios, particularly those involving high-resolution or pose-stable domains, highlighting the importance of strong inductive biases over architectural novelty alone. Further, to characterize robustness quantitatively, we introduced the <i>Unified Robustness Metric</i> (URM), which integrates the average cross-domain degradation performance into a single score. A qualitative saliency analysis also corroborated the numerical findings by exposing over-confidence and contextual bias in misclassifications. Overall, this study suggests that challenges in gender recognition are much more evident in cross-domain settings than under the commonly reported within-domain context. Finally, we formalize an open evaluation protocol that can serve as a baseline for future works of this kind.https://www.mdpi.com/1424-8220/25/13/4161gender recognitionpedestrian attribute recognitionsoft biometricsmodel generalizationcomputer visiondeep learning
spellingShingle Nzakiese Mbongo
Kailash A. Hambarde
Hugo Proença
From One Domain to Another: The Pitfalls of Gender Recognition in Unseen Environments
Sensors
gender recognition
pedestrian attribute recognition
soft biometrics
model generalization
computer vision
deep learning
title From One Domain to Another: The Pitfalls of Gender Recognition in Unseen Environments
title_full From One Domain to Another: The Pitfalls of Gender Recognition in Unseen Environments
title_fullStr From One Domain to Another: The Pitfalls of Gender Recognition in Unseen Environments
title_full_unstemmed From One Domain to Another: The Pitfalls of Gender Recognition in Unseen Environments
title_short From One Domain to Another: The Pitfalls of Gender Recognition in Unseen Environments
title_sort from one domain to another the pitfalls of gender recognition in unseen environments
topic gender recognition
pedestrian attribute recognition
soft biometrics
model generalization
computer vision
deep learning
url https://www.mdpi.com/1424-8220/25/13/4161
work_keys_str_mv AT nzakiesembongo fromonedomaintoanotherthepitfallsofgenderrecognitioninunseenenvironments
AT kailashahambarde fromonedomaintoanotherthepitfallsofgenderrecognitioninunseenenvironments
AT hugoproenca fromonedomaintoanotherthepitfallsofgenderrecognitioninunseenenvironments