Human Pose and Shape Estimation (HPSE) from RGB images can be broadly
categorized into two main groups: parametric and non-parametric approaches.
Parametric techniques leverage a low-dimensional statistical body model for
realistic results, whereas recent non-parametric methods achieve higher
precision by directly regressing the 3D coordinates of the human body. Despite
their strengths, both approaches face limitations: the parameters of
statistical body models pose challenges as regression targets, and predicting
3D coordinates introduces computational complexities and issues related to
smoothness. In this work, we take a novel approach to address the HPSE problem.
We introduce a unique method involving a low-dimensional discrete latent
representation of the human mesh, framing HPSE as a classification task.
Instead of predicting body model parameters or 3D vertex coordinates, our focus
is on forecasting the proposed discrete latent representation, which can be
decoded into a registered human mesh. This innovative paradigm offers two key
advantages: firstly, predicting a low-dimensional discrete representation
confines our predictions to the space of anthropomorphic poses and shapes;
secondly, by framing the problem as a classification task, we can harness the
discriminative power inherent in neural networks. Our proposed model, VQ-HPS, a
transformer-based architecture, forecasts the discrete latent representation of
the mesh, trained through minimizing a cross-entropy loss. Our results
demonstrate that VQ-HPS outperforms the current state-of-the-art non-parametric
approaches while yielding results as realistic as those produced by parametric
methods. This highlights the significant potential of the classification
approach for HPSE.

通过引入一种低维离散潜在表示的独特方法，我们将人体姿态和形状估计 (Human Pose and Shape Estimation，HPSE) 问题作为一个分类任务来解决，预测离散潜在表示，从而实现对人体网格的编码。我们的模型 VQ-HPS 在训练中通过最小化交叉熵损失实现，结果显示它在 HPSE 问题上胜过当前最先进的无参数方法，并且产生与参数方法一样逼真的结果，这突显了分类方法在 HPSE 中的重要潜力。