DmsFIQA: Dms-Specific Face Image Quality Assessment for In-Cabin Driver Monitoring System

Qiushi Guo, Zhiye Lin, Yuanqing Luo, Lin Luo, Jason Rambach

Keywords: FIQA, DMS

TL;DR: We propose DmsFIQA, a DMS-specific face quality model with near-annotation-free supervision that reliably ranks in-cabin faces and improves downstream DMS face recognition by filtering low-quality frames.

Abstract: Face Image Quality Assessment (FIQA) is a critical preprocessing step for face-related applications such as face recognition and face anti-spoofing. However, most prior FIQA methods are developed for generic capture conditions and do not transfer well to domain-specific settings such as in-cabin Driver Monitoring Systems (DMS), where strong domain shift arises from frequent occlusions, large pose variations, challenging illumination, and partial-face captures. In this paper, we introduce DmsFIQA, a DMS-oriented FIQA framework that fills this gap. We construct a DMS face-quality dataset covering diverse real-world conditions and design a two-stage annotation pipeline that minimizes manual labeling: (i) we first obtain coarse quality estimates via large-scale model–based automatic assessment, and (ii) we refine the supervision by ranking images through identity-consistent similarity between per-identity query images and high-quality templates. We evaluate DmsFIQA on both FIQA prediction and downstream DMS face recognition. Experiments show that DmsFIQA produces more fine-grained and reliable quality estimates and effectively filters low-quality faces, leading to improved robustness of the overall DMS recognition pipeline.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 10