Face alignment: improving the accuracy of fast models using domain-specific unlabelled data and a teacher–student scheme

Constantino Alvarez Casado, Miguel Bordallo Lopez

Research output: Contribution to journalArticleScientificpeer-review

Abstract

Face alignment is a crucial step in multiple face analysis and recognition tasks. The current state-of-the-art is comprised by very slow methods based on deep learning that require computationally heavy inference and very fast methods based on cascades of regressors that lack the ability to cope with complicated cases or extreme poses. The authors show how collecting a small subset of unlabelled domain-specific data can improve the accuracy of fast-inference models utilising data annotated by a slower one and a teacher–student architecture. In the proposed solution, they annotate a small subset of facial images belonging to two challenging domains using a slow but more accurate model, and this data is used to incrementally train a fast one. Their results show that by adding as little as a 5% of challenging data, they can reduce the error rate in a specific domain up to 30% without losing any generalisation abilities. This training scheme has applicability in numerous computer vision and engineering problems where computational power and model size are constrained by the application and platform or real-time operation is a requirement.
Original languageEnglish
Pages (from-to)646 - 648
Number of pages3
JournalElectronics Letters
Volume55
Issue number11
DOIs
Publication statusPublished - 2019
MoE publication typeA1 Journal article-refereed

    Fingerprint

Keywords

  • computer vision
  • learning (artificial intelligence)
  • face recognition
  • engineering problems
  • computational power
  • model size
  • numerous computer vision
  • training scheme
  • generalisation abilities
  • specific domain
  • challenging data
  • slow but more accurate model
  • challenging domains
  • teacher–student architecture
  • fast-inference models
  • unlabelled domain-specific data
  • extreme poses
  • complicated cases
  • fast methods
  • computationally heavy inference
  • deep learning
  • slow methods
  • current state-of-the-art
  • recognition tasks
  • multiple face analysis
  • crucial step
  • teacher–student scheme
  • domain-specific unlabelled data
  • fast models
  • face alignment

Cite this