Fig. 13 | Scientific Reports

Fig. 13

From: Enhanced spatiotemporal skeleton modeling: integrating part-joint attention with dynamic graph convolution

Fig. 13

Radar chart visualization of input granularity and motion feature ablation results. Each axis represents one prediction horizon (80–400 ms), and the values denote MPJPE (mm). Models compared include joint-only, part-only, joint+part, and full model with velocity and acceleration features. The figure illustrates that combining joint and part-level features, along with motion dynamics, leads to consistently lower errors across all horizons.

Back to article page