Improving Domain Generalization in 3D Human Pose Estimation Using a Dual-Augmentation Approach

EasyChair Preprint 14634

Abstract

This article explores a dual-augmentation approach designed to improve domain generalization in 3D human pose estimation. Domain generalization is a critical challenge in computer vision, especially in 3D pose estimation, where models trained on specific datasets often fail to generalize well to new environments. The proposed approach integrates geometric and photometric augmentations to create diverse training samples, enhancing the model's robustness and performance across unseen domains. Through extensive experiments on benchmark datasets, the study demonstrates that the dual-augmentation approach significantly reduces pose estimation errors and increases accuracy compared to traditional single-augmentation methods. The findings suggest that this approach offers a promising solution for improving domain generalization in computer vision tasks.

Keyphrases: 3D human pose estimation, Domain Generalization, Geometric Augmentation, Photometric Augmentation, data augmentation

@booklet{EasyChair:14634,
  author    = {Adeoye Ibrahim},
  title     = {Improving Domain Generalization in 3D Human Pose Estimation Using a Dual-Augmentation Approach},
  howpublished = {EasyChair Preprint 14634},
  year      = {EasyChair, 2024}}