Octuplet Loss: Make Face Recognition Robust to Image Resolution

TL;DR

This paper introduces octuplet loss, a novel training method that enhances face recognition robustness across different image resolutions by leveraging relationships between high- and low-resolution images.

Contribution

The authors propose octuplet loss, a new loss function that improves cross-resolution face verification performance by fine-tuning existing models with synthetic down-sampled images.

Findings

Significantly improves cross-resolution face verification accuracy.

Achieves 95.12% accuracy on XQLFW and 99.73% on LFW datasets.

Enhances low-to-low resolution face verification performance.

Abstract

Image resolution, or in general, image quality, plays an essential role in the performance of today's face recognition systems. To address this problem, we propose a novel combination of the popular triplet loss to improve robustness against image resolution via fine-tuning of existing face recognition models. With octuplet loss, we leverage the relationship between high-resolution images and their synthetically down-sampled variants jointly with their identity labels. Fine-tuning several state-of-the-art approaches with our method proves that we can significantly boost performance for cross-resolution (high-to-low resolution) face verification on various datasets without meaningfully exacerbating the performance on high-to-high resolution images. Our method applied on the FaceTransformer network achieves 95.12% face verification accuracy on the challenging XQLFW dataset while reaching 99.73% on the LFW database. Moreover, the low-to-low face verification accuracy benefits from our method. We release our code to allow seamless integration of the octuplet loss into existing frameworks.