$\text{Di}^2\text{Pose}$: Discrete Diffusion Model for Occluded 3D Human Pose Estimation

TL;DR

Di2Pose introduces a discrete diffusion framework for occluded 3D human pose estimation, effectively reducing unrealistic pose generation and improving understanding of occlusions by discretizing pose representations and modeling them in latent space.

Contribution

The paper proposes a novel discrete diffusion model for occluded 3D human pose estimation that enhances physical plausibility and occlusion understanding through pose quantization and latent space diffusion.

Findings

Outperforms existing methods on benchmarks like Human3.6M and 3DPW.

Effectively handles occlusions and generates realistic 3D poses.

Reduces search space to physically plausible configurations.

Abstract

Continuous diffusion models have demonstrated their effectiveness in addressing the inherent uncertainty and indeterminacy in monocular 3D human pose estimation (HPE). Despite their strengths, the need for large search spaces and the corresponding demand for substantial training data make these models prone to generating biomechanically unrealistic poses. This challenge is particularly noticeable in occlusion scenarios, where the complexity of inferring 3D structures from 2D images intensifies. In response to these limitations, we introduce the Discrete Diffusion Pose ($\text{Di}^2\text{Pose}$), a novel framework designed for occluded 3D HPE that capitalizes on the benefits of a discrete diffusion model. Specifically, $\text{Di}^2\text{Pose}$ employs a two-stage process: it first converts 3D poses into a discrete representation through a \emph{pose quantization step}, which is subsequently modeled in latent space through a \emph{discrete diffusion process}. This methodological innovation restrictively confines the search space towards physically viable configurations and enhances the model's capability to comprehend how occlusions affect human pose within the latent space. Extensive evaluations conducted on various benchmarks (e.g., Human3.6M, 3DPW, and 3DPW-Occ) have demonstrated its effectiveness.