论文阅读 [TPAMI-2022] Locally Connected Network for Monocular 3D Human Pose Estimation

论文阅读 [TPAMI-2022] Locally Connected Network for Monocular 3D Human Pose Estimation

论文搜索(studyai.com)

搜索论文: Locally Connected Network for Monocular 3D Human Pose Estimation

搜索论文: http://www.studyai.com/search/whole-site/?q=Locally+Connected+Network+for+Monocular+3D+Human+Pose+Estimation

关键字(Keywords)

Three-dimensional displays; Two dimensional displays; Pose estimation; Feature extraction; Solid modeling; Training; Task analysis; 3D human pose estimation; locally connected network; graph convolution

机器学习; 机器视觉

姿态估计; 动作识别; 图卷积网络; 三维人体

摘要(Abstract)

We present an approach for 3D human pose estimation from monocular images.

提出了一种基于单目图像的三维人体姿态估计方法。.

The approach consists of two steps: it first estimates a 2D pose from an image and then estimates the corresponding 3D pose.

该方法包括两个步骤：首先从图像中估计二维姿势，然后估计相应的三维姿势。.

This paper focuses on the second step.

本文的重点是第二步。.

Graph convolutional network (GCN) has recently become the de facto standard for human pose related tasks such as action recognition.

图形卷积网络（GCN）最近已成为动作识别等与人体姿势相关任务的事实标准。.

However, in this work, we show that GCN has critical limitations when it is used for 3D pose estimation due to the inherent weight sharing scheme.

然而，在这项工作中，我们发现，由于固有的权重共享方案，GCN在用于3D姿势估计时存在严重的局限性。.

The limitations are clearly exposed through a novel reformulation of GCN, in which both GCN and Fully Connected Network (FCN) are its special cases.

通过对GCN的一种新的重新表述，可以清楚地看到其局限性，其中GCN和完全连接网络（FCN）都是它的特例。.

In addition, on top of the formulation, we present locally connected network (LCN) to overcome the limitations of GCN by allocating dedicated rather than shared filters for different joints.

此外，在该公式的基础上，我们提出了局部连接网络（LCN），通过为不同的关节分配专用而非共享的过滤器来克服GCN的局限性。.

We jointly train the LCN network with a 2D pose estimator such that it can handle inaccurate 2D poses.

我们将LCN网络与2D姿势估计器联合训练，使其能够处理不精确的2D姿势。.

We evaluate our approach on two benchmark datasets and observe that LCN outperforms GCN, FCN, and the state-of-the-art methods by a large margin.

我们在两个基准数据集上评估了我们的方法，并观察到LCN在很大程度上优于GCN、FCN和最先进的方法。.

More importantly, it demonstrates strong cross-dataset generalization ability because of sparse connections among body joints…

更重要的是，由于人体关节之间的稀疏连接，它展示了强大的跨数据集泛化能力。。.

作者(Authors)

[‘Hai Ci’, ‘Xiaoxuan Ma’, ‘Chunyu Wang’, ‘Yizhou Wang’]