MyoMechanix: Biomechanically-Grounded Compositional Skilled Activity Understanding and Coaching
The paper introduces a multimodal benchmark that pairs exercise motion with muscle activity for finer coaching feedback.
MyoMechanix covers more than 7,500 samples across 20 weight-loaded actions from 38 subjects. The dataset synchronizes multiview RGB video, 3D pose, sEMG, and other physiological signals. The authors add a Fitness Knowledge Graph linking actions, phases, errors, and corrective feedback, then use CUBIST to attribute mistakes and generate interpretable assessments. Experiments report gains from multimodal sensing and structured reasoning, including state-of-the-art results for CUBIST and a Video2EMG task meant to infer muscle signals from video. ArXiv · AI/CL/LG's note
MyoMechanix covers more than 7,500 samples across 20 weight-loaded actions from 38 subjects. The dataset synchronizes multiview RGB video, 3D pose, sEMG, and other physiological signals. The authors add a Fitness Knowledge Graph linking actions, phases, errors, and corrective feedback, then use CUBIST to attribute mistakes and generate interpretable assessments. Experiments report gains from multimodal sensing and structured reasoning, including state-of-the-art results for CUBIST and a Video2EMG task meant to infer muscle signals from video. ArXiv · AI/CL/LG's note
score 4