Related Experiment Video
Updated: Jul 1, 2026
06:45
Automated Joint Space Detection Improves Bone Segmentation Accuracy
Published on: November 28, 2025
CVDII: Enhancing One-Shot Skeleton Action Recognition Through Cross-View Dynamic Information Interaction.
Summary
This study introduces a new Cross-View Dynamic Information Interaction (CVDII) framework to improve one-shot 3D skeleton action recognition by managing diverse action styles. CVDII effectively balances shared and discriminative information for better feature separation and recognition accuracy.
Area of Science:
- Computer Vision
- Machine Learning
- Artificial Intelligence
Background:
- One-shot 3D skeleton action recognition faces challenges due to diverse intra-class execution styles.
- Excessive discriminative information hinders the creation of separable feature spaces.
Purpose of the Study:
- To mitigate the over-influence of discriminative information in action recognition.
- To leverage shared information among intra-class action executions for improved recognition.
Main Methods:
- Proposed the dynamic information interaction module (DIIM) to manage shared and discriminative information.
- Developed a guided evolution pool within DIIM for retrieving shared execution information.
- Devised a shared-discriminative projection strategy (SDPS) for targeted information mining from different skeleton data views.
Main Results:
- The Cross-View Dynamic Information Interaction (CVDII) framework integrates DIIM and SDPS.
- CVDII effectively addresses discriminative information redundancy caused by varied action execution styles.
- Experiments on NTU 60, NTU 120, PKU-MMD, and Kinetics datasets show remarkable performance.
Conclusions:
- The proposed CVDII framework significantly enhances one-shot 3D skeleton action recognition.
- Effective management of shared and discriminative information is key to overcoming execution style diversity.
- The method demonstrates robust performance across multiple benchmark datasets.
