IG-FIQA: Improving Face Image Quality Assessment through Intra-class Variance Guidance robust to Inaccurate Pseudo-Labels
This work addresses a specific issue in FIQA for face recognition systems, representing an incremental improvement over existing methods.
The paper tackled the problem of inaccurate pseudo-labels in face image quality assessment (FIQA) by introducing IG-FIQA, which uses intra-class variance guidance to mitigate adverse effects from low-variance classes, achieving novel state-of-the-art performance on benchmark datasets.
In the realm of face image quality assesment (FIQA), method 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 this method. To address this issue, we present 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. On various benchmark datasets, our proposed method, IG-FIQA, achieved novel state-of-the-art (SOTA) performance.