Large-Scale Video-Based Activated Muscle Group Estimation in Unconstrained Environments

This work introduces a new large-scale video-based dataset, MuscleMap, for estimating activated muscle groups during physical activities in unconstrained environments. The proposed TRANSM3E model leverages multi-classification tokens, cross-modality knowledge distillation, and fusion mechanisms to achieve state-of-the-art performance with superior generalizability compared to existing approaches.