Video Title:
Supplementary Video 1. Temporal Visualization of Joint-Level Attention Across SHREC’22 Gesture Classes

Video Description:
This video provides a real-time visualization of joint-level attention for all gesture classes in the SHREC’22 dataset. Attention scores were derived using Grad-CAM applied to SKETCH-based representations and mapped back to 3D joint coordinates. The color intensity of each joint reflects the magnitude of attention, progressing from blue (low attention) to red (high attention). This visualization highlights how the model dynamically focuses on semantically meaningful joints over time, offering insights into its interpretability and gesture recognition behavior.

