Human Pose Estimation with Spatial Contextual Information
Hong Zhang1 Hao Ouyang2 Shu Liu3 Xiaojuan Qi5 Xiaoyong Shen3 Ruigang Yang1,4 Jiaya Jia3,5 1Baidu Research, Baidu Inc. 2Hong Kong University of Science and Technology 3YouTu Lab, Tencent 4 University of Kentucky 5The Chinese University of Hong Kong
Abstract
We explore the importance of spatial contextual information in human pose estimation. Most state-of-the-art pose networks are trained in a multi-stage manner and produce several auxiliary predictions for deep supervision. With this principle, we present two conceptually simple and yet computational efficient modules, namely Cascade Prediction Fusion (CPF) and Pose Graph Neural Network (PGNN), to exploit underlying contextual information. Cascade prediction fusion accumulates prediction maps from previous stages to extract informative signals. The resulting maps also function as a prior to guide prediction at following stages. To promote spatial correlation among joints, our PGNN learns a structured representation of human pose as a graph. Direct message passing between different joints is enabled and spatial relation is captured. These two modules require very limited computational complexity. Experimental results demonstrate that our method consistently outperforms previous methods on MPII and LSP benchmark.
中文速览
人体姿态估计（human pose estimation）需要准确定位身体各关节点，而现有方法往往忽视了不同关节之间的空间上下文关联——比如左膝盖的位置可以帮助推断被遮挡的左脚踝在哪里。为此，研究者提出了两个轻量级模块：级联预测融合（Cascade Prediction Fusion, CPF）和姿态图神经网络（Pose Graph Neural Network, PGNN）。CPF将多阶段预测网络中每一阶段生成的"热力图"逐级积累并作为先验信息传递给下一阶段，使粗略定位信号和高层语义信息都能被充分利用；PGNN则把每个关节点建模为图上的节点，通过图神经网络中的消息传递机制让各关节并行地互相"通信"，从而捕获关节间的结构关系，避免了传统概率图模型顺序更新容易累积误差的问题。在MPII和LSP两个主流基准数据集上，这两个模块的组合以很低的额外计算代价刷新了当时的最优结果，证明显式建模空间上下文关系对提升姿态估计精度确实行之有效。
原文 arXiv:1901.01760;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1901.01760v1