ADG-Pose: Automated Dataset Generation for Real-World Human Pose Estimation

TL;DR

ADG-Pose is a novel automated dataset generation method that creates customized datasets reflecting real-world challenges like occlusion and distance, significantly improving human pose estimation accuracy in complex scenes.

Contribution

The paper introduces ADG-Pose, a new automated dataset generation approach that enhances training data for real-world human pose estimation under challenging conditions.

Findings

20% accuracy increase in moderate distance and occlusion scenes

4X improvement in distant scenes where previous models failed

Models trained with ADG-Pose outperform those trained on traditional datasets

Abstract

Recent advancements in computer vision have seen a rise in the prominence of applications using neural networks to understand human poses. However, while accuracy has been steadily increasing on State-of-the-Art datasets, these datasets often do not address the challenges seen in real-world applications. These challenges are dealing with people distant from the camera, people in crowds, and heavily occluded people. As a result, many real-world applications have trained on data that does not reflect the data present in deployment, leading to significant underperformance. This article presents ADG-Pose, a method for automatically generating datasets for real-world human pose estimation. These datasets can be customized to determine person distances, crowdedness, and occlusion distributions. Models trained with our method are able to perform in the presence of these challenges where those trained on other datasets fail. Using ADG-Pose, end-to-end accuracy for real-world skeleton-based action recognition sees a 20% increase on scenes with moderate distance and occlusion levels, and a 4X increase on distant scenes where other models failed to perform better than random.