The Blessing and the Curse of the Noise behind Facial Landmark Annotations

TL;DR

This paper investigates how annotation noise affects facial landmark detection stability and accuracy, proposing metrics and solutions to model and mitigate noise impacts in training neural networks.

Contribution

It introduces two metrics for landmark stability, models annotation noise in datasets, and explores noise effects with solutions to improve detection performance.

Findings

Improved landmark stability and accuracy.

Quantitative measurement of annotation noise influence.

Effective noise mitigation strategies for training neural networks.

Abstract

The evolving algorithms for 2D facial landmark detection empower people to recognize faces, analyze facial expressions, etc. However, existing methods still encounter problems of unstable facial landmarks when applied to videos. Because previous research shows that the instability of facial landmarks is caused by the inconsistency of labeling quality among the public datasets, we want to have a better understanding of the influence of annotation noise in them. In this paper, we make the following contributions: 1) we propose two metrics that quantitatively measure the stability of detected facial landmarks, 2) we model the annotation noise in an existing public dataset, 3) we investigate the influence of different types of noise in training face alignment neural networks, and propose corresponding solutions. Our results demonstrate improvements in both accuracy and stability of detected facial landmarks.