Updates

This page features condensed results of all the analysis that was done. Introduction Fairness evaluation studies across Biometric Modalities Fairness in Face Recognition Ocular : Motivation Research Objectives Covariates to be Studied Datasets Used Gender Classification Updates to Training Across Model - VIS Across Model - NIR Across Model - NIR (Cross Dataset) Across Databalance - VIS Across Databalance - NIR Across Race - VIS Across Age - VIS Across Modality - VIS Across Modality - NIR Across Device - VIS Across Lighting - VIS Across Spectrum Subject Verification Across Model - VIS Across Model - NIR Across Databalance - VIS Across Databalance - NIR Across Race - VIS Across Modality - VIS Across Modality - NIR Across Device - VIS Across Lighting - VIS Across Spectrum Summary

Training Datasets VISOB - VIS, Gender, Subject, Race, Device, Lighting, Synthetic ND-IRIS - NIR, Gender, Subject UFPR - VIS, Gender, Subject, Age Notredame - NIR, VIS, Face, Ocular, Gender, Subject, Race Polyuiris - NIR, VIS, Subject Casia-Lamp - NIR, Subject Casia-Thousand - NIR, Subject FFHQ-Aging - VIS, Face, Ocular, Gender Casia-Syn - NIR, Subject, Synthetic Evaluation Datasets MOBIUS - VIS, Subject, Device, Light, Gaze, Quality, Gender, Age, Color VISA-IRIS - VIS, Subject, Gender, Face, Ocular IMP - NIR, VIS, Subject I-SOCIAL-DB - VIS, Subject Cross-Eyed - NIR, VIS, Subject UNAB - NIR, Subject, Gender, Makeup, Contacts Subject Verification Across Model - VIS Experiment ID Gender GMean GSTD IMean ISTD Sensitivity index (d') AUC J-Index MCC EER ZeroFMR FMR1000 FMR100 FMR20 FMR10 ZeroFNMR efficientnet_b0 male 0.

Introduction Fairness evaluation studies across Biometric Modalities Fairness in Face Recognition Ocular : Motivation Research Objectives Covariates to be Studied Datasets Used Gender Classification Across Models Across Models on UFPR Periocular - RGB Across Models on Notredame Ocular - NIR Across Varying Data Balances and Gender UFPR - RGB Notredame - NIR Across Age UFPR - RGB TABLE I: Accuracy of CNN-based Gender Classification on Left Ocular Region among Young (18 to 39 years), Middle (40 to 59 years) and Older Adults (60 to 79 years).

Gender Classification Across Race on Ocular Training and Testing Dataset used - VISOB + Notredame Test Data is Limited and Imbalanced TABLE I: Gender Classification results across race groups - Ocular Datasets.

Dataset Gender Classification Results Subject Verification Results Subject Verification All Results Table Visualizations GradCAM GradCAM Guided GradCAM Guided GradCAM Occlusion Sensitivity Occlusion Sensitivity Vanilla Gradients Vanilla Gradients Integrated Gradients Integrated Gradients SmoothGrad -- Gradients x Input Gradients x Input Averaged Map × References Bibliography called, but no references

Dataset The UFPR-Periocular dataset was created to obtain images in unconstrained scenarios that contain realistic noises caused by occlusion, blur, and variations in lighting, distance, and angles. The gender distribution of the subjects is $(53,65\%)$ male and $(46,35\%)$ female, and approximately $66\%$ of the subjects are under $31$ years old.

Dataset Dataset includes 33,660 ocular samples from 1,122 subjects captured by 196 different mobile. All images were resized to224×224. the complete dataset is classified into three age-groups: namely young (18 to 39), middle-aged (40 to 59), and older adults (60 to 79).