Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cross-Age Contrastive Learning for Age-Invariant Face Recognition

Haoyi Wang, Victor Sanchez, Chang-Tsun Li

TL;DR

The paper tackles age-invariant face recognition (AIFR) under limited cross-age data by introducing CACon, a semi-supervised contrastive framework that incorporates a third, age-synthesized sample to enforce identity consistency across ages. It extends contrastive learning to a triplet setting and proposes a triplet NT-Xent loss, enabling the model to maximize agreement among features from the same subject across different ages. Evaluations on FG-NET, MORPH II, and CACD-VS show that CACon achieves state-of-the-art performance in both homogeneous and cross-dataset scenarios, outperforming strong baselines like SimCLR. The approach demonstrates robust cross-age generalization and highlights the practicality of combining synthetic age augmentation with contrastive learning for age-invariant biometric recognition.

Abstract

Cross-age facial images are typically challenging and expensive to collect, making noise-free age-oriented datasets relatively small compared to widely-used large-scale facial datasets. Additionally, in real scenarios, images of the same subject at different ages are usually hard or even impossible to obtain. Both of these factors lead to a lack of supervised data, which limits the versatility of supervised methods for age-invariant face recognition, a critical task in applications such as security and biometrics. To address this issue, we propose a novel semi-supervised learning approach named Cross-Age Contrastive Learning (CACon). Thanks to the identity-preserving power of recent face synthesis models, CACon introduces a new contrastive learning method that leverages an additional synthesized sample from the input image. We also propose a new loss function in association with CACon to perform contrastive learning on a triplet of samples. We demonstrate that our method not only achieves state-of-the-art performance in homogeneous-dataset experiments on several age-invariant face recognition benchmarks but also outperforms other methods by a large margin in cross-dataset experiments.

Cross-Age Contrastive Learning for Age-Invariant Face Recognition

TL;DR

The paper tackles age-invariant face recognition (AIFR) under limited cross-age data by introducing CACon, a semi-supervised contrastive framework that incorporates a third, age-synthesized sample to enforce identity consistency across ages. It extends contrastive learning to a triplet setting and proposes a triplet NT-Xent loss, enabling the model to maximize agreement among features from the same subject across different ages. Evaluations on FG-NET, MORPH II, and CACD-VS show that CACon achieves state-of-the-art performance in both homogeneous and cross-dataset scenarios, outperforming strong baselines like SimCLR. The approach demonstrates robust cross-age generalization and highlights the practicality of combining synthetic age augmentation with contrastive learning for age-invariant biometric recognition.

Abstract

Cross-age facial images are typically challenging and expensive to collect, making noise-free age-oriented datasets relatively small compared to widely-used large-scale facial datasets. Additionally, in real scenarios, images of the same subject at different ages are usually hard or even impossible to obtain. Both of these factors lead to a lack of supervised data, which limits the versatility of supervised methods for age-invariant face recognition, a critical task in applications such as security and biometrics. To address this issue, we propose a novel semi-supervised learning approach named Cross-Age Contrastive Learning (CACon). Thanks to the identity-preserving power of recent face synthesis models, CACon introduces a new contrastive learning method that leverages an additional synthesized sample from the input image. We also propose a new loss function in association with CACon to perform contrastive learning on a triplet of samples. We demonstrate that our method not only achieves state-of-the-art performance in homogeneous-dataset experiments on several age-invariant face recognition benchmarks but also outperforms other methods by a large margin in cross-dataset experiments.
Paper Structure (10 sections, 6 equations, 2 figures, 3 tables)

This paper contains 10 sections, 6 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Cross-Age Contrastive Learning
  3. Problem Formulation
  4. Data Augmentation
  5. Modified Contrastive Loss
  6. Experiments
  7. Datasets
  8. Experiment Settings
  9. Comparison with State-of-the-Art Methods
  10. Conclusion

Figures (2)

  • Figure 1: Data augmentation strategy used in (a) conventional contrastive learning, where two augmented samples are used to learn the shared features representing the identity within the input image and (b) CACon, where an additional sample is generated by a face synthesis model and used to learn the common identity features across different ages.
  • Figure 2: Architecture of the proposed CACon where the black arrows indicate the conventional contrastive learning and the red arrows illustrate the additional path used to learn age-invariant identity features.