Spatial–temporal graph attention networks for skeleton-based action recognition

Huang, Qingqing; Zhou, Fengyu; He, JiaKai; Zhao, Yang; Qin, Runze

doi:10.1117/1.jei.29.5.053003

Cited by 9 publications

(4 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Because the Fourier transform usually assumes that the graph structure is fixed, in the frequency domain, it is difficult for the graph convolutional network to directly process the dynamic graph because the graph changes over time, and involves the eigenvalue decomposition of the matrix, resulting in high complex computation. Therefore, spatial domains-based graph convolutional network is usually used for action recognition [21][22][23][24].…”

Section: Related Workmentioning

confidence: 99%

“…However, the manually defined topology is difficult to achieve the aggregation of nodes without physical connection, which greatly affects the recognition rate and generalization of the network. To obtain more connectivity relationships between the joints, and improve the expressive ability of the model, ST-GAT [21] defines the spatiotemporal adjacent nodes and aggregation functions of the root node through the attention mechanism. The AS-GCN [22] introduces an encoder-decoder structure to capture action-specific potential dependencies directly from the actions.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Multi-Scale Adaptive Graph Convolution Network for Skeleton-Based Action Recognition

Hu,

Fang,

Han

et al. 2024

IEEE Access

View full text Add to dashboard Cite

The skeleton-based action recognition technology can effectively avoid the background interference and occlusion problems in the image. However, the recognition of similar actions is still a challenge. In this paper, a multi-scale dynamic topological modeling method (MDTM) is proposed to solve this problem. The topological modeling through the convolution kernel generated from the original data, increases the connection of the convolution process to the original data compared to the previous randomly generated ones, effectively distinguishing similar actions. In addition, MDTM uses a multi-scale temporal convolutional network to obtain a wider receptive field, which can effectively extract the temporal information in the action. At the same time, a dynamic topology learning method is utilized to design a spatiotemporal information extractor that can effectively extract the spatiotemporal information in the action to dynamically adjust the topology structure. Extensive experiments have performed on three large-scale datasets, NTU RGB + D 60, NTU RGB + D 120, and NW-UCLA to validate the effectiveness of MDTM. The results show that MART-GCN performs better than the others in terms of accuracy and number of parameters.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Multi-Scale Adaptive Graph Convolution Network for Skeleton-Based Action Recognition

Hu,

Fang,

Han

et al. 2024

IEEE Access

View full text Add to dashboard Cite

show abstract

“…(4) When designing graph convolution kernels, the strategy does not take into account the relationship between the number of kernels and the complexity of the graph structure. As a result, the topology of the network will be too smooth and will not be able to fully capture the distance between distant joints while the learning process is taking place [10][11][12][13][14][15][16].…”

Section: Introductionmentioning

confidence: 99%

Big Data System for Dragon Boat Rowing Action Training Based on Multidimensional Stereo Vision

Tao

2022

Mathematical Problems in Engineering

View full text Add to dashboard Cite

With the rapid advancement of artificial intelligence technology and the widespread use of sensing technology in education, human-computer interaction teaching has gradually developed in sports and education. Traditionally, teachers explain and demonstrate the fundamentals of movements first, then organize exercises, and students gradually consolidate technical movements through repetition. This process requires teachers to repeatedly explain, such that students can develop movement concepts, and to assist students in correcting their movements through practice. Eventually, students can master the dragon boat’s paddling movements. Teachers frequently struggle to observe all of their students’ movements and are therefore unable to correct them in a timely and effective manner. To address the aforementioned issues, this paper proposes a big data system for multidimensional stereo vision training in dragon boat paddling action. The stock price action recognition of dragon boat paddles and the movement of students’ dragon boat paddles are realized through the multidimensional fusion of attention mechanism and spatiotemporal graph convolution. Make judgments to more effectively guide and train students’ paddling movements.

show abstract

“…Considering this problem, to achieve state-of-the-art performance, we propose the use of Graph Convolutional Neural Networks, and we create a novel gating mechanism applied to Semantic Graph Convolutions(SGCs) [15] named Semantic Graph Attention (SGAT). We enhance the analysis of global correlation, which is crucial for understanding human actions [16]. Our new layer can learn both channel-wise weights for edges, combine them with kernel matrices, and features inter-dependencies over channels.…”

Section: Introductionmentioning

confidence: 99%

Semantic graph attention networks and tensor decompositions for computer vision and computer graphics

Schirmer

Lopes²,

Velho³

2021

Anais Estendidos Da XXXIV Conference on Graphics, Patterns and Images (SIBRAPI Estendido 2021)

View full text Add to dashboard Cite

This thesis proposes new architectures for deep neural networks with attention enhancement and multilinear algebra methods to increase their performance. We also explore graph convolutions and their particularities. We focus here on the problems related to real-time human pose estimation. We explore different architectures to reduce computational complexity, and, as a result, we propose two novel neural network models for 2D and 3D pose estimation. We also introduce a new architecture for Graph attention networks called Semantic Graph Attention.

show abstract

Spatial–temporal graph attention networks for skeleton-based action recognition

Cited by 9 publications

References 0 publications

Multi-Scale Adaptive Graph Convolution Network for Skeleton-Based Action Recognition

Multi-Scale Adaptive Graph Convolution Network for Skeleton-Based Action Recognition

Big Data System for Dragon Boat Rowing Action Training Based on Multidimensional Stereo Vision

Semantic graph attention networks and tensor decompositions for computer vision and computer graphics

Contact Info

Product

Resources

About