Joint Training of a Convolutional Network and a Graphical Model for Human Pose Estimation
Jonathan Tompson, Arjun Jain, Yann LeCun, Christoph Bregler New York University {tompson, ajain, yann,
Abstract
This paper proposes a new hybrid architecture that consists of a deep Convolutional Network and a Markov Random Field. We show how this architecture is successfully applied to the challenging problem of articulated human pose estimation in monocular images. The architecture can exploit structural domain constraints such as geometric relationships between body joint locations. We show that joint training of these two model paradigms improves performance and allows us to significantly outperform existing state-of-the-art techniques.
中文速览
人体姿态估计（human pose estimation）长期面临一个难题：深度卷积网络（Convolutional Network）虽然能高效检测各个关节的位置，却很难自动学到"肩膀必须靠近脸部"这类人体结构约束，导致预测结果中出现大量解剖学上不合理的误检。为此，本文提出将卷积网络关节检测器与马尔可夫随机场（Markov Random Field, MRF）空间模型结合成一个统一架构：前者输出每个关节的热力图（heat-map），后者通过可学习的卷积先验对关节间的空间关系建模，近似实现置信传播（loopy belief propagation），并以端到端方式与检测器联合训练。在FLIC和LSP两个公开基准上，这一联合框架显著优于当时所有已有方法，证明将结构先验以可反向传播的方式嵌入深度网络，能有效消除解剖学错误的异常预测，大幅提升高精度区间的定位性能。
原文 arXiv:1406.2984;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1406.2984v2