NSTL国家科技图书文献中心

1. Towards Physical World Backdoor Attacks Against Skeleton Action Recognition NSTL国家科技图书文献中心

Qichen Zheng | Yi Yu ... - 《Computer Vision - ECCV 2024,Part XLVIII》 - European Conference on Computer Vision - 2025, - 215～233 - 共19页

摘要：Skeleton Action Recognition (SAR) has | introduce the Physical Skeleton* Backdoor Attacks (PSBA* | actions as triggers into the original skeleton* data. By* | misclassify any skeleton* sequences into the target class* | representation of the human skeletal* structure. Despite its*

关键词： Backdoor attacks | Skeleton action recognition

2. Language-Assisted Skeleton Action Understanding for Skeleton-Based Temporal Action Segmentation NSTL国家科技图书文献中心

Haoyu Ji | Bowen Chen ... - 《Computer Vision - ECCV 2024,Part LIV》 - European Conference on Computer Vision - 2025, - 400～417 - 共18页

摘要：Skeleton-based Temporal Action Segmentation | skeleton action understanding. In this work, we propose a | Language-assisted Skeleton* Action Understanding (LaSA* | actions in long, untrimmed skeletal* motion sequences* | (STAS) aims to densely segment and classify human

关键词： Video understanding | Skeleton-based action segmentation | Language-Assisted learning | Attention | Contrastive learning

3. Dynamic Skeleton Association Transformer for Dyadic Interaction Action Recognition NSTL国家科技图书文献中心

Zixian Liu | Longfei Zhang ... - 《Pattern Recognition and Computer Vision,Part VII》 - Chinese Conference on Pattern Recognition and Computer Vision - 2025, - 554～569 - 共16页

摘要： skeleton data as graphs, it has always been the primary | method for skeleton-based human action recognition | . However, when dealing with interaction skeleton | dynamically updating the connections between the skeleton | can well focus on the correlation between skeleton

关键词： Skeleton-based interaction recognition | Action recognition | Graph convolutional networks | Transformer

4. MacDiff: Unified Skeleton Modeling with Masked Conditional Diffusion NSTL国家科技图书文献中心

Lehong Wu | Lilang Lin ... - 《Computer Vision - ECCV 2024,Part XXVI》 - European Conference on Computer Vision - 2025, - 110～128 - 共19页

摘要： for skeleton-based human action understanding | skeleton modeling. For the first time, we leverage | diffusion models as effective skeleton* representation* | skeletons with spa-cial sparsity and temporal redundancy | of skeletons. Furthermore, we theoretically

关键词： Self-supervised learning | Unified skeleton modeling | Diffusion model

5. Dribbling Posture Estimation Using a Human Skeleton Model NSTL国家科技图书文献中心

Masaya Endo | Yoshinari Kameda - 《International Workshop on Advanced Imaging Technology (IWAIT) 2025》 - International Workshop on Advanced Imaging Technology - 2025, - 135101G.1～135101G.6 - 共6页

摘要： human skeleton. To train this model, we handcrafted a | videos of players who were dribbling with a skeleton | skeletons holding the ball, which we have named the | existing skeletal* estimation model, OpenPose, and propose* | We aim to evaluate dribbling movements in

关键词： Human posture estimation | Soccer | Articulated posture model | Human skeleton model

6. Multi‑scale spatiotemporal topology unveiled: enhancing skeleton‑based action recognition NSTL国家科技图书文献中心

Hongwei Chen | Jianpeng Wang ... - 《The Journal of Supercomputing》 - 2025,81(1) - 10.1～10.26 - 共26页

摘要：In recent years, skeleton-based action | joint movements play a critical role in skeleton-based | robustness of human skeletons* in complex environments* | recognition has received considerable attention due to the | . However, many existing methods face challenges in

关键词： Action recognition | Multi-scale | Human skeleton | 3D CNN

7. Learning by Aligning 2D Skeleton Sequences and Multi-modality Fusion NSTL国家科技图书文献中心

Quoc-Huy Tran | Muhammad Ahmed ... - 《Computer Vision - ECCV 2024,Part L》 - European Conference on Computer Vision - 2025, - 141～161 - 共21页

摘要： CASA, where sequences of 3D skeleton* coordinates are* | sequences of 2D skeleton* heatmaps as input. Unlike CASA* | only, we feed 2D skeleton* heatmaps to a video* | . Finally, fusing 2D skeleton* heatmaps with RGB videos* | utilize 2D skeleton* heatmap inputs and the first to*

关键词： Temporal video alignment | Temporal 2D skeleton sequence alignment | Multi-modality fusion | Self-supervised learning

8. Continuous Human Activity Recognition System by Using Human Skeleton Information NSTL国家科技图书文献中心

Wenbang Dou | Aulia Saputra Azhar ... - 《Intelligent Robotics and Applications,Part X》 - International Conference on Intelligent Robotics and Applications - 2025, - 3～17 - 共15页

摘要： System that generates a 3D human skeleton* model and* | human skeleton* model using depth information obtained* | In recent years, as society continues to face | an aging population, there is a shortage of | caregivers, leading to an increase in the demand for

关键词： Human skeleton model | Human activity recognition | Continuous learning

9. Language-guided temporal primitive modeling for skeleton-based action recognition NSTL国家科技图书文献中心

Pan Q. | Xie X. - 《Neurocomputing》 - 2025,613(Jan.14) - 1.1～1.12 - 共12页

摘要： skeleton-based action recognition primarily focus on | language-guided skeleton* representation learning network* | guide the skeleton* feature learning from the ordered* | primitive guidance aligns the features of skeleton* clips* | internal structure within the action from skeletons. To

关键词： Language guidance | Large language model | Skeleton-based action recognition | Temporal primitive

10. Idempotent Unsupervised Representation Learning for Skeleton-Based Action Recognition NSTL国家科技图书文献中心

Lilang Lin | Lehong Wu ... - 《Computer Vision - ECCV 2024,Part XXVI》 - European Conference on Computer Vision - 2025, - 75～92 - 共18页

摘要： recognition tasks. However, in skeleton-based action | theory and methodology and propose a novel skeleton | Generative models, as a powerful technique for | generation, also gradually become a critical tool for | recognition, the features obtained from existing pre-trained

摘要：Skeleton Action Recognition (SAR) has | introduce the Physical Skeleton* Backdoor Attacks (PSBA* | actions as triggers into the original skeleton* data. By* | misclassify any skeleton* sequences into the target class* | representation of the human skeletal* structure. Despite its*

关键词： Backdoor attacks | Skeleton action recognition

摘要：Skeleton-based Temporal Action Segmentation | skeleton action understanding. In this work, we propose a | Language-assisted Skeleton* Action Understanding (LaSA* | actions in long, untrimmed skeletal* motion sequences* | (STAS) aims to densely segment and classify human

关键词： Video understanding | Skeleton-based action segmentation | Language-Assisted learning | Attention | Contrastive learning

摘要： skeleton data as graphs, it has always been the primary | method for skeleton-based human action recognition | . However, when dealing with interaction skeleton | dynamically updating the connections between the skeleton | can well focus on the correlation between skeleton

关键词： Skeleton-based interaction recognition | Action recognition | Graph convolutional networks | Transformer

摘要： for skeleton-based human action understanding | skeleton modeling. For the first time, we leverage | diffusion models as effective skeleton* representation* | skeletons with spa-cial sparsity and temporal redundancy | of skeletons. Furthermore, we theoretically

关键词： Self-supervised learning | Unified skeleton modeling | Diffusion model

摘要： human skeleton. To train this model, we handcrafted a | videos of players who were dribbling with a skeleton | skeletons holding the ball, which we have named the | existing skeletal* estimation model, OpenPose, and propose* | We aim to evaluate dribbling movements in

关键词： Human posture estimation | Soccer | Articulated posture model | Human skeleton model

摘要：In recent years, skeleton-based action | joint movements play a critical role in skeleton-based | robustness of human skeletons* in complex environments* | recognition has received considerable attention due to the | . However, many existing methods face challenges in

关键词： Action recognition | Multi-scale | Human skeleton | 3D CNN

摘要： CASA, where sequences of 3D skeleton* coordinates are* | sequences of 2D skeleton* heatmaps as input. Unlike CASA* | only, we feed 2D skeleton* heatmaps to a video* | . Finally, fusing 2D skeleton* heatmaps with RGB videos* | utilize 2D skeleton* heatmap inputs and the first to*

关键词： Temporal video alignment | Temporal 2D skeleton sequence alignment | Multi-modality fusion | Self-supervised learning

摘要： System that generates a 3D human skeleton* model and* | human skeleton* model using depth information obtained* | In recent years, as society continues to face | an aging population, there is a shortage of | caregivers, leading to an increase in the demand for

关键词： Human skeleton model | Human activity recognition | Continuous learning

摘要： skeleton-based action recognition primarily focus on | language-guided skeleton* representation learning network* | guide the skeleton* feature learning from the ordered* | primitive guidance aligns the features of skeleton* clips* | internal structure within the action from skeletons. To

关键词： Language guidance | Large language model | Skeleton-based action recognition | Temporal primitive

摘要： recognition tasks. However, in skeleton-based action | theory and methodology and propose a novel skeleton | Generative models, as a powerful technique for | generation, also gradually become a critical tool for | recognition, the features obtained from existing pre-trained

关键词： Self-supervised learning | Skeleton-based action recognition | Contrastive learning

4008-161-200 800-990-8900

国家科技图书文献中心

摘要：Skeleton Action Recognition (SAR) has | introduce the Physical Skeleton Backdoor Attacks (PSBA | actions as triggers into the original skeleton data. By | misclassify any skeleton sequences into the target class | representation of the human skeletal structure. Despite its

关键词： Backdoor attacks | Skeleton action recognition

摘要：Skeleton-based Temporal Action Segmentation | skeleton action understanding. In this work, we propose a | Language-assisted Skeleton Action Understanding (LaSA | actions in long, untrimmed skeletal motion sequences | (STAS) aims to densely segment and classify human

关键词： Video understanding | Skeleton-based action segmentation | Language-Assisted learning | Attention | Contrastive learning

摘要： skeleton data as graphs, it has always been the primary | method for skeleton-based human action recognition | . However, when dealing with interaction skeleton | dynamically updating the connections between the skeleton | can well focus on the correlation between skeleton

关键词： Skeleton-based interaction recognition | Action recognition | Graph convolutional networks | Transformer

摘要： for skeleton-based human action understanding | skeleton modeling. For the first time, we leverage | diffusion models as effective skeleton representation | skeletons with spa-cial sparsity and temporal redundancy | of skeletons. Furthermore, we theoretically

关键词： Self-supervised learning | Unified skeleton modeling | Diffusion model

摘要： human skeleton. To train this model, we handcrafted a | videos of players who were dribbling with a skeleton | skeletons holding the ball, which we have named the | existing skeletal estimation model, OpenPose, and propose | We aim to evaluate dribbling movements in

关键词： Human posture estimation | Soccer | Articulated posture model | Human skeleton model

摘要：In recent years, skeleton-based action | joint movements play a critical role in skeleton-based | robustness of human skeletons in complex environments | recognition has received considerable attention due to the | . However, many existing methods face challenges in

关键词： Action recognition | Multi-scale | Human skeleton | 3D CNN

摘要： CASA, where sequences of 3D skeleton coordinates are | sequences of 2D skeleton heatmaps as input. Unlike CASA | only, we feed 2D skeleton heatmaps to a video | . Finally, fusing 2D skeleton heatmaps with RGB videos | utilize 2D skeleton heatmap inputs and the first to

关键词： Temporal video alignment | Temporal 2D skeleton sequence alignment | Multi-modality fusion | Self-supervised learning

摘要： System that generates a 3D human skeleton model and | human skeleton model using depth information obtained | In recent years, as society continues to face | an aging population, there is a shortage of | caregivers, leading to an increase in the demand for

关键词： Human skeleton model | Human activity recognition | Continuous learning

摘要： skeleton-based action recognition primarily focus on | language-guided skeleton representation learning network | guide the skeleton feature learning from the ordered | primitive guidance aligns the features of skeleton clips | internal structure within the action from skeletons. To

关键词： Language guidance | Large language model | Skeleton-based action recognition | Temporal primitive

摘要： recognition tasks. However, in skeleton-based action | theory and methodology and propose a novel skeleton | Generative models, as a powerful technique for | generation, also gradually become a critical tool for | recognition, the features obtained from existing pre-trained

关键词： Self-supervised learning | Skeleton-based action recognition | Contrastive learning

4008-161-200 800-990-8900

国家科技图书文献中心

摘要：Skeleton Action Recognition (SAR) has | introduce the Physical Skeleton* Backdoor Attacks (PSBA* | actions as triggers into the original skeleton* data. By* | misclassify any skeleton* sequences into the target class* | representation of the human skeletal* structure. Despite its*

摘要：Skeleton-based Temporal Action Segmentation | skeleton action understanding. In this work, we propose a | Language-assisted Skeleton* Action Understanding (LaSA* | actions in long, untrimmed skeletal* motion sequences* | (STAS) aims to densely segment and classify human

摘要： for skeleton-based human action understanding | skeleton modeling. For the first time, we leverage | diffusion models as effective skeleton* representation* | skeletons with spa-cial sparsity and temporal redundancy | of skeletons. Furthermore, we theoretically

摘要： human skeleton. To train this model, we handcrafted a | videos of players who were dribbling with a skeleton | skeletons holding the ball, which we have named the | existing skeletal* estimation model, OpenPose, and propose* | We aim to evaluate dribbling movements in

摘要：In recent years, skeleton-based action | joint movements play a critical role in skeleton-based | robustness of human skeletons* in complex environments* | recognition has received considerable attention due to the | . However, many existing methods face challenges in

摘要： CASA, where sequences of 3D skeleton* coordinates are* | sequences of 2D skeleton* heatmaps as input. Unlike CASA* | only, we feed 2D skeleton* heatmaps to a video* | . Finally, fusing 2D skeleton* heatmaps with RGB videos* | utilize 2D skeleton* heatmap inputs and the first to*

摘要： System that generates a 3D human skeleton* model and* | human skeleton* model using depth information obtained* | In recent years, as society continues to face | an aging population, there is a shortage of | caregivers, leading to an increase in the demand for

摘要： skeleton-based action recognition primarily focus on | language-guided skeleton* representation learning network* | guide the skeleton* feature learning from the ordered* | primitive guidance aligns the features of skeleton* clips* | internal structure within the action from skeletons. To