Multiscale Graph Redefining: Correlation-Based Multiscale Graph Clustering Network for Human Motion Prediction.

Area of Science:

• Computer Vision
• Machine Learning
• Human Motion Analysis

Background:

• Graph Convolutional Networks (GCNs) are effective for 3-D skeleton-based human motion prediction.
• Existing methods use fixed multiscale graphs, which overlook dynamic semantic information and exhibit unstable joint correlations.
• This limitation hinders accurate human motion prediction.

Purpose of the Study:

• To develop an adaptive multiscale graph representation learning network for improved 3-D human motion prediction.
• To address the limitations of fixed multiscale graphs in capturing dynamic motion characteristics.
• To enhance the accuracy and robustness of human motion prediction models.

Main Methods:

• Introduced a novel correlation-based multiscale graph clustering network (CMGC).
• Employed adaptive multiscale graph generation and selective restoration for diverse motion feature extraction.
• Integrated discrete wavelet transform (DWT) to mitigate signal loss from discrete cosine transform (DCT) domain modeling.

Main Results:

• CMGC achieved superior performance on Human 3.6M, CMU Mocap, and 3DPW datasets, outperforming state-of-the-art methods by an average of 11.2% in 3-D mean per joint position error (MPJPE).
• Demonstrated a 6.5% reduction in mean angle error (MAE) on the Human3.6M dataset compared to previous approaches.
• The adaptive multiscale graph representation learning significantly enhanced prediction accuracy.

Conclusions:

• The proposed CMGC network effectively captures dynamic motion correlations through adaptive multiscale graph learning.
• CMGC represents a significant advancement in 3-D skeleton-based human motion prediction accuracy and robustness.
• The findings suggest a promising direction for future research in human motion analysis and prediction.