Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mitigating Longitudinal Performance Degradation in Child Face Recognition Using Synthetic Data

Afzal Hossain, Stephanie Schuckers

TL;DR

This study tackles longitudinal degradation in child face recognition caused by rapid growth by testing synthetic data as a stabilizing augmentation. Using MagFace embeddings on the YFA dataset with an identity-disjoint protocol, the authors compare pretrained, real-only fine-tuned, and real-plus-synthetic fine-tuned models. They show that synthetic augmentation, when leakage-free and restricted to training identities, markedly improves recognition stability across enrollment–verification gaps up to 36 months, reducing EER from over 1%–2% to sub-1% levels. The work demonstrates a risk-aware, practical approach to enhancing pediatric identity persistence, while outlining safeguards and directions for future generalization and fairness analyses.

Abstract

Longitudinal face recognition in children remains challenging due to rapid and nonlinear facial growth, which causes template drift and increasing verification errors over time. This work investigates whether synthetic face data can act as a longitudinal stabilizer by improving temporal robustness of child face recognition models. Using an identity disjoint protocol on the Young Face Aging (YFA) dataset, we evaluate three settings: (i) pretrained MagFace embeddings without dataset specific fine-tuning, (ii) MagFace fine-tuned using authentic training faces only, and (iii) MagFace fine-tuned using a combination of authentic and synthetically generated training faces. Synthetic data is generated using StyleGAN2 ADA and incorporated exclusively within the training identities; a post generation filtering step is applied to mitigate identity leakage and remove artifact affected samples. Experimental results across enrollment verification gaps from 6 to 36 months show that synthetic-augmented fine tuning substantially reduces error rates relative to both the pretrained baseline and real only fine tuning. These findings provide a risk aware assessment of synthetic augmentation for improving identity persistence in pediatric face recognition.

Mitigating Longitudinal Performance Degradation in Child Face Recognition Using Synthetic Data

TL;DR

This study tackles longitudinal degradation in child face recognition caused by rapid growth by testing synthetic data as a stabilizing augmentation. Using MagFace embeddings on the YFA dataset with an identity-disjoint protocol, the authors compare pretrained, real-only fine-tuned, and real-plus-synthetic fine-tuned models. They show that synthetic augmentation, when leakage-free and restricted to training identities, markedly improves recognition stability across enrollment–verification gaps up to 36 months, reducing EER from over 1%–2% to sub-1% levels. The work demonstrates a risk-aware, practical approach to enhancing pediatric identity persistence, while outlining safeguards and directions for future generalization and fairness analyses.

Abstract

Longitudinal face recognition in children remains challenging due to rapid and nonlinear facial growth, which causes template drift and increasing verification errors over time. This work investigates whether synthetic face data can act as a longitudinal stabilizer by improving temporal robustness of child face recognition models. Using an identity disjoint protocol on the Young Face Aging (YFA) dataset, we evaluate three settings: (i) pretrained MagFace embeddings without dataset specific fine-tuning, (ii) MagFace fine-tuned using authentic training faces only, and (iii) MagFace fine-tuned using a combination of authentic and synthetically generated training faces. Synthetic data is generated using StyleGAN2 ADA and incorporated exclusively within the training identities; a post generation filtering step is applied to mitigate identity leakage and remove artifact affected samples. Experimental results across enrollment verification gaps from 6 to 36 months show that synthetic-augmented fine tuning substantially reduces error rates relative to both the pretrained baseline and real only fine tuning. These findings provide a risk aware assessment of synthetic augmentation for improving identity persistence in pediatric face recognition.
Paper Structure (8 sections, 4 figures, 1 table)

This paper contains 8 sections, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. Dataset Description
  5. Feature Extraction
  6. Training and Evaluation Protocol
  7. Experimental Results
  8. Conclusion and Future Work

Figures (4)

  • Figure 1: Example synthetic face images illustrating the outcome of the quality filtering process. The top row presents visually consistent samples retained for training, while the bottom row shows artifact affected images that were excluded from the training set.
  • Figure 2: False Reject Rate (FRR) as a function of False Accept Rate (FAR) across increasing enrollment–recognition time gaps, evaluated using pretrained MagFace (no fine-tuning).
  • Figure 3: False Reject Rate (FRR) as a function of False Accept Rate (FAR) across increasing enrollment–recognition time gaps, evaluated using MagFace fine-tuned on authentic (real-only) data.
  • Figure 4: False Reject Rate (FRR) as a function of False Accept Rate (FAR) across increasing enrollment–recognition time gaps, evaluated using MagFace fine-tuned on authentic + synthetic data.