LIAAD: Lightweight Attentive Angular Distillation for Large-scale Age-Invariant Face Recognition

TL;DR

LIAAD is a novel lightweight distillation method that improves large-scale age-invariant face recognition by transferring knowledge from heavy networks to a compact model, demonstrating robustness and high accuracy on multiple datasets.

Contribution

The paper introduces LIAAD, a lightweight attentive angular distillation approach that effectively transfers knowledge from heavy networks to a small model for large-scale age-invariant face recognition.

Findings

LIAAD achieves high accuracy on large-scale face recognition datasets.

The approach is robust against age variations and scalable to large datasets.

LIAAD outperforms prior methods in efficiency and accuracy.

Abstract

Disentangled representations have been commonly adopted to Age-invariant Face Recognition (AiFR) tasks. However, these methods have reached some limitations with (1) the requirement of large-scale face recognition (FR) training data with age labels, which is limited in practice; (2) heavy deep network architectures for high performance; and (3) their evaluations are usually taken place on age-related face databases while neglecting the standard large-scale FR databases to guarantee robustness. This work presents a novel Lightweight Attentive Angular Distillation (LIAAD) approach to Large-scale Lightweight AiFR that overcomes these limitations. Given two high-performance heavy networks as teachers with different specialized knowledge, LIAAD introduces a learning paradigm to efficiently distill the age-invariant attentive and angular knowledge from those teachers to a lightweight student network making it more powerful with higher FR accuracy and robust against age factor. Consequently, LIAAD approach is able to take the advantages of both FR datasets with and without age labels to train an AiFR model. Far apart from prior distillation methods mainly focusing on accuracy and compression ratios in closed-set problems, our LIAAD aims to solve the open-set problem, i.e. large-scale face recognition. Evaluations on LFW, IJB-B and IJB-C Janus, AgeDB and MegaFace-FGNet with one million distractors have demonstrated the efficiency of the proposed approach on light-weight structure. This work also presents a new longitudinal face aging (LogiFace) database \footnote{This database will be made available} for further studies in age-related facial problems in future.