Core Concepts
Proposing a novel GCN-DevLSTM network for skeleton-based action recognition, leveraging path development to enhance temporal dynamics.
Abstract
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.
Stats
"Our method(J) outperforms 10.4% and 9.7% on NTU-120 dataset compared to Logsig-RNN."
Quotes
"Our method(J) outperforms 10.4% and 9.7% on NTU-120 dataset compared to Logsig-RNN."