Adversarial PoseNet: A Structure-aware Convolutional Network for Human Pose Estimation

TL;DR

This paper introduces a structure-aware convolutional network for human pose estimation that leverages adversarial training to incorporate geometric priors, improving accuracy in occluded or overlapping scenarios.

Contribution

It proposes a novel adversarial framework that implicitly encodes human body structure priors into pose estimation, addressing limitations of explicit constraint learning.

Findings

Improved pose estimation accuracy in occlusion scenarios

Effective incorporation of geometric priors via adversarial training

Enhanced plausibility of predicted human poses

Abstract

For human pose estimation in monocular images, joint occlusions and overlapping upon human bodies often result in deviated pose predictions. Under these circumstances, biologically implausible pose predictions may be produced. In contrast, human vision is able to predict poses by exploiting geometric constraints of joint inter-connectivity. To address the problem by incorporating priors about the structure of human bodies, we propose a novel structure-aware convolutional network to implicitly take such priors into account during training of the deep network. Explicit learning of such constraints is typically challenging. Instead, we design discriminators to distinguish the real poses from the fake ones (such as biologically implausible ones). If the pose generator (G) generates results that the discriminator fails to distinguish from real ones, the network successfully learns the priors.