IG-FIQA: Improving Classifiability-Based Face Image Quality Assessment Through Intra-Class Variance Guidance

Abstract

In the realm of face image quality assessment (FIQA), methods based on sample relative classification have shown impressive performance. However, the quality scores used as pseudo-labels assigned from images of classes with low intra-class variance could be unrelated to the actual quality in such methods. To address this issue, we present intra-class variance guidance for FIQA (IG-FIQA), a novel approach to guide FIQA training, introducing a weight parameter to alleviate the adverse impact of these classes. This method involves estimating sample intra-class variance at each iteration during training, ensuring minimal computational overhead and straightforward implementation. Furthermore, this paper proposes an on-the-fly data augmentation methodology for improved generalization performance in FIQA. Across various benchmark datasets, our proposed method, IG-FIQA, achieved notable accuracy improvements compared to conventional state-of-the-art (SOTA) FIQA methods and ensures stable performance in face recognition systems.