ViTNT-FIQA: Training-Free Face Image Quality Assessment with Vision Transformers

2026-01-09

arXiv.org

AI Summary

This work proposes a training-free, single forward-pass method for face image quality assessment that overcomes the limitations of existing approaches, which often rely on multiple forward or backward passes and utilize only the final-layer features of Vision Transformers (ViTs). The proposed method uniquely evaluates quality by analyzing the evolutionary stability of patch embeddings across intermediate ViT layers: high-quality images exhibit smooth feature transitions between layers, whereas low-quality ones show irregular fluctuations. By computing and aggregating Euclidean distances between L2-normalized embeddings of consecutive Transformer blocks, the method generates an image-level quality score using any off-the-shelf pretrained ViT without fine-tuning. It achieves state-of-the-art performance across eight benchmarks, including LFW and IJB-C, offering both high efficiency and plug-and-play versatility.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Qualitative Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Face Image Quality Assessment (FIQA) is essential for reliable face recognition systems. Current approaches primarily exploit only final-layer representations, while training-free methods require multiple forward passes or backpropagation. We propose ViTNT-FIQA, a training-free approach that measures the stability of patch embedding evolution across intermediate Vision Transformer (ViT) blocks. We demonstrate that high-quality face images exhibit stable feature refinement trajectories across blocks, while degraded images show erratic transformations. Our method computes Euclidean distances between L2-normalized patch embeddings from consecutive transformer blocks and aggregates them into image-level quality scores. We empirically validate this correlation on a quality-labeled synthetic dataset with controlled degradation levels. Unlike existing training-free approaches, ViTNT-FIQA requires only a single forward pass without backpropagation or architectural modifications. Through extensive evaluation on eight benchmarks (LFW, AgeDB-30, CFP-FP, CALFW, Adience, CPLFW, XQLFW, IJB-C), we show that ViTNT-FIQA achieves competitive performance with state-of-the-art methods while maintaining computational efficiency and immediate applicability to any pre-trained ViT-based face recognition model.

Problem

• Face Image Quality Assessment
• Training-Free
• Vision Transformers
• Patch Embedding Stability
• Computational Efficiency

Innovation

• training-free
• Vision Transformer
• face image quality assessment
• patch embedding stability
• single forward pass