Improving Cross-Domain Human Pose Estimation Without Source Data Access: Additional Details

EasyChair Preprint 14633

Abstract

Human pose estimation is a challenging task in computer vision, with applications ranging from augmented reality to human-computer interaction. Traditional methods often rely on access to source domain data to achieve accurate pose estimation. However, in cross-domain settings, source data may not be available due to privacy concerns, security regulations, or logistical limitations. This article explores advanced methodologies to enhance human pose estimation across different domains without relying on source data. We delve into various techniques, including domain adaptation, transfer learning, and synthetic data generation, to mitigate the absence of source data. The findings presented in this supplementary information section are aimed at providing a deeper understanding of how these techniques can be effectively employed to overcome challenges posed by cross-domain scenarios. By offering a comprehensive evaluation and analysis, we demonstrate that these methods significantly improve the performance of pose estimation models, even when source data is inaccessible.

Keyphrases: Cross-Domain Human Pose Estimation, Domain Adaptation, Pose estimation models, Synthetic Data Generation, Transfer Learning, adversarial training, computer vision, domain shift, privacy-preserving machine learning, self-supervised learning

@booklet{EasyChair:14633,
  author    = {Kayode Sheriffdeen},
  title     = {Improving Cross-Domain Human Pose Estimation Without Source Data Access: Additional Details},
  howpublished = {EasyChair Preprint 14633},
  year      = {EasyChair, 2024}}