Spatial Focus Attention for Fine-Grained Skeleton-Based Action Tasks

Comparative Study on Binomial Distribution Content in High School Textbooks

“…In order to solve this problem, the third approach uses GCNs (Yan et al, 2018;Shi et al, 2019b;Ye et al, 2020;Zhang et al, 2020a;Shi et al, 2019a;Kong et al, 2022;Gao et al, 2021;Liu et al, 2022;Peng et al, 2021;Song et al, 2020;Wu et al, 2021;Liu et al, 2021) to capture topological graph structure of skeleton. For example, Yan et al (Yan et al, 2018) proposed spatial-temporal graph convolutional networks (ST-GCN) to extract topological spatial-temporal features, where a static graph is used to capture relationship among joints.…”

“…Compared with the conventional RGB video, 3D skeleton owning high-level representation is light-weight and robust to both view differences and complicated background. Therefore, 3D skeleton based action recognition has been widely investigated with methods based on handcrafted features (Weng et al, 2017;Xia et al, 2012), Convolutional Neural Networks (CNNs) (Ke et al, 2017a,b;Li et al, 2017;Hou et al, 2018;Li et al, 2019a;Xu et al, 2018;, Recurrent Neural Networks (RNNs) (Li et al, 2018(Li et al, , 2019bLiu et al, 2018;Song et al, 2017) and Graph Convolutional Networks (GCNs) (Yan et al, 2018;Shi et al, 2019b;Ye et al, 2020;Zhang et al, 2020a;Shi et al, 2019a;Kong et al, 2022;Gao et al, 2021;Liu et al, 2022;Peng et al, 2021). However, these methods are developed in a fully supervised manner and require extensive annotated labels, which is expensive and time-consuming.…”

“…Putting aside their individual problems such as designing relevant task and differentiating positive and negative pairs, both approaches suffer from severe overfitting. The existing networks used in supervised learning (Shi et al, 2019b;Ye et al, 2020;Zhang et al, 2020a;Shi et al, 2019a;Kong et al, 2022;Gao et al, 2021;Liu et al, 2022;Peng et al, 2021) cannot work effectively in the unsupervised learning due to this severe overfitting. Consequently, the existing unsupervised learning methods employ very simple models, either using only basic RNN models or using very small models with fewer neurons (Zhang et al, 2022b,a).…”

3sG: Three‐stage guidance for indoor human action recognition

“…For example, obscured hands or legs, this type of motion information is easily lost. For the skeleton-based approaches [10][11][12][13][14][15][16], lightweight skeleton data can make the model computationally less costly, but it is easy to misunderstand similar motion trajectory of the action without image visual information.…”