Основные понятия
Proposing a novel GCN-DevLSTM network for skeleton-based action recognition, leveraging path development to enhance temporal dynamics.
Аннотация
The content introduces the GCN-DevLSTM model for skeleton-based action recognition, emphasizing the importance of capturing temporal dynamics. It discusses the challenges in current models and presents empirical studies demonstrating the superiority of the proposed hybrid model. The paper also includes comparisons with state-of-the-art methods, robustness analysis, and ablation studies to highlight the effectiveness of the DevLSTM module.
-
Introduction
- Importance of skeleton-based action recognition.
- Challenges in designing effective spatio-temporal representations.
-
GCN-DevLSTM Model
- Introduction of DevLSTM module for capturing temporal relationships.
- Integration with GCN for spatial correlations.
-
Data Extraction
- "Our method(J) outperforms 10.4% and 9.7% on NTU-120 dataset compared to Logsig-RNN."
-
Comparison with State-of-the-Art Methods
- Superior performance of GCN-DevLSTM over existing approaches on NTU datasets.
-
Robustness Analysis
- Demonstrates robustness to missing frames, outperforming other methods consistently.
-
Ablation Studies
- Path development layer emerges as a primary contributor to performance improvement.
Статистика
"Our method(J) outperforms 10.4% and 9.7% on NTU-120 dataset compared to Logsig-RNN."
Цитаты
"Our method(J) outperforms 10.4% and 9.7% on NTU-120 dataset compared to Logsig-RNN."