Context Modeling in 3D Human Pose Estimation: A Unified Perspective

Ma, Xiaoxuan; Su, Jiajun; Wang, Qianqian; Ci, Hai; Wang, Yizhou

doi:10.1109/cvpr46437.2021.00617

Cited by 47 publications

(15 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Since human joints can be naturally seemed as a graph structure, many works [2,6,23,31,35,36,53] try to utilize this information. Zhao et al [53] take the human joints as the nodes of graph and then build a semantic graph convolutional network to learn joints relation.…”

Section: Graph Reasoning In Pose Estimationmentioning

confidence: 99%

“…Zhao et al [53] take the human joints as the nodes of graph and then build a semantic graph convolutional network to learn joints relation. More works try to learn a better graph representation for human pose, such as spatial-temporal graph convolutional network [2], dynamic graph convolutional network [36], context pose [23] and so on. Other works [6,52] study the local or global structure of the human body.…”

Section: Graph Reasoning In Pose Estimationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

“…For 3D human pose estimation, the common way to decode 3D pose is based on heatmaps [23,28,41,43,54], which decodes 3D coordinates from 3D heatmaps by an isolated structure as Figure 1 (a). To mitigate occlusion problem, [23,54] model context information in coordinate level and feature level, respectively. Recent works [31,55] bridge these two levels to some extent by estimating the 3D keypoints offsets, directly decoding keypoints coordinates from root point with 3D offsets.…”

Section: Introductionmentioning

confidence: 99%

“…This mechanism enables the best decoding paths for each target joint. Although some works [23,52,53] use graph convolutional network or local human structure to learn better structural features, the learned body graph is also fixed. In the inference phase, this learned structural graph can't be adjusted adaptively with the occlusion of the input image.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Dynamic Graph Reasoning for Multi-person 3D Pose Estimation

Qiu¹,

Qian-Sheng²,

Wang³

et al. 2022

Preprint

View full text Add to dashboard Cite

Multi-person 3D pose estimation is a challenging task because of occlusion and depth ambiguity, especially in the cases of crowd scenes. To solve these problems, most existing methods explore modeling body context cues by enhancing feature representation with graph neural networks or adding structural constraints. However, these methods are not robust for their single-root formulation that decoding 3D poses from a root node with a pre-defined graph. In this paper, we propose GR-M3D, which models the Multi-person 3D pose estimation with dynamic Graph Reasoning. The decoding graph in GR-M3D is predicted instead of pre-defined. In particular, It firstly generates several data maps and enhances them with a scale and depth aware refinement module (SDAR). Then multiple root keypoints and dense decoding paths for each person are estimated from these data maps. Based on them, dynamic decoding graphs are built by assigning path weights to the decoding paths, while the path weights are inferred from those enhanced data maps. And this process is named dynamic graph reasoning (DGR). Finally, the 3D poses are decoded according to dynamic decoding graphs for each detected person. GR-M3D can adjust the structure of the decoding graph implicitly by adopting soft path weights according to input data, which makes the decoding graphs be adaptive to different input persons to the best extent and more capable of handling occlusion and depth ambiguity than previous methods. We empirically show that the proposed bottom-up approach even outperforms top-down methods and achieves state-of-the-art results on three 3D pose datasets. CCS CONCEPTS• Computing methodologies → Object recognition.

show abstract