Optimizing Network Structure for 3D Human Pose Estimation

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

doi:10.1109/iccv.2019.00235

Cited by 198 publications

(149 citation statements)

References 21 publications

Supporting

Mentioning

148

Contrasting

Order By: Relevance

“…Especially, we choose a threshold of 150mm in the calculation of PCK. Table 3 shows the quantitative results of different scenes from which our results are closest to the method of [27]. In outdoor scenes, our result is 5.7% higher than Chang et al [44].…”

Section: ) For Mpi-inf-3dhpmentioning

confidence: 60%

“…On average our approach reduces estimation error by 6.8 millimeters under protocol-I, compared to the method [13]. We are also aware of the fact that our method will not outperform these methods based on the weight nonsharing mechanism in graphs [27]. Because weight nonsharing mechanism enhances the model's complexity and representation capability.…”

Section: ) For Human 36mmentioning

confidence: 91%

“…Recent researches have achieved state-of-the-art performance in modeling the relations of visual temporal sequences [23], [24] and visual objects [25], [26] with GCNs. Ci et al [27] present a locally connected network to improve the representation capability of GCNs. Liu et al [28] first systematically analyze the mechanism of weight sharing in graphs.…”

Section: B Graph Convolutional Networkmentioning

confidence: 99%

“…By adding a refinement network, the Model 5 reduces the estimation error by 0.9mm compared [44] is a method with data augmentation. Algorithm [27] is based on weighted non-sharing mechanism. * denotes the methods without capturing spatial structure information.…”

Section: E Ablation Studymentioning

confidence: 99%

See 3 more Smart Citations

3D Human Pose Estimation With Spatial Structure Information

Huang

Tang

2021

IEEE Access

View full text Add to dashboard Cite

Estimating 3D human poses from 2D poses is a challenging problem due to joints selfocclusion, weak generalization, and inherent ambiguity of recovering depth. Actually, there exists spatial structure dependence on human body key points which can be used to alleviate the problem of joints selfocclusion. Therefore, we represent human pose as a directed graph and propose a network implemented with graph convolution to predict 3D poses from the given 2D poses. In the digraph, we determine the connection weight of each edge according to the error distribution of joints estimation. This makes our model robust to noise. By optimizing coarse 3D estimation and adversarial learning, our algorithm can successfully improve the accuracy of estimation and relieve the ambiguity of mapping. Through testing on Human 3.6M and MPI-INF-3DHP datasets, we achieve excellent quantitative performance. More importantly, our algorithm also has a superior generalization to outdoor dataset MPII by the pre-training process.INDEX TERMS 3D human poses, graph convolutional networks, adversarial learning, geometric priors, gradient vanish, in-the-wild scenes.

show abstract

Section: ) For Mpi-inf-3dhpmentioning

confidence: 60%

Section: ) For Human 36mmentioning

confidence: 91%

Section: B Graph Convolutional Networkmentioning

confidence: 99%

Section: E Ablation Studymentioning

confidence: 99%

See 2 more Smart Citations

3D Human Pose Estimation With Spatial Structure Information

Huang

Tang

2021

IEEE Access

View full text Add to dashboard Cite

show abstract

“…And the approaches either utilize example-based re-finement or rely on very strong assumption including scaled orthographic cameras, or calibrated perspective cameras [4], [5]. Recently, inspired by the success of deep convolutional neural network (DCNN), lots of DCNN-based 3D human pose estimation methods are also proposed [6]- [10]. The 3D pose estimation methods can be roughly categorized into two kinds: one-step model and two-steps model.…”

Section: Introductionmentioning

confidence: 99%

An Adaptive Viewpoint Transformation Network for 3D Human Pose Estimation

et al. 2020

View full text Add to dashboard Cite

Human pose estimation from a monocular image has attracted lots of interest due to its huge potential application in many areas. The performance of 2D human pose estimation has been improved a lot with the emergence of deep convolutional neural network. In contrast, the recovery of 3D human pose from an 2D pose is still a challenging problem. Currently, most of the methods try to learn a universal map, which can be applied for all human poses in any viewpoints. However, due to the large variety of human poses and camera viewpoints, it is very difficult to learn a such universal mapping from current datasets for 3D pose estimation. Instead of learning a universal map, we propose to learn an adaptive viewpoint transformation module, which transforms the 2D human pose to a more suitable viewpoint for recovering the 3D human pose. Specifically, our transformation module takes a 2D pose as input and predicts the transformation parameters. Rather than some hand-crafted criteria, this module is directly learned from the datasets and depends on the input 2D pose in testing phrase. Then the 3D pose is recovered from this transformed 2D pose. Since the difficulty of 3D pose recovery becomes smaller, we can obtain more accurate estimation results. Experiments on Human3.6M and MPII datasets show that the proposed adaptive viewpoint transformation can improve the performance of 3D human pose estimation. INDEX TERMS 3D human pose estimation, adaptive viewpoint transformation, deep convolutional neural network

show abstract