Can Computer Vision Systems Infer Your Muscle Activity From Video? Meet Muscles In Action (mia): A New Dataset To Learn To Incorporate Muscle Activity Into Human Motion Representations

Researchers from Columbia University have developed a new dataset called “Muscles in Action” (MIA) which includes 12.5 hours of synchronized video and surface electromyography (sEMG) data. This dataset captures ten subjects performing various exercises and is used to develop a representation that can predict muscle activation from video and reconstruct human motion from muscle activation data. The primary aim is to comprehend the complex connection between the underlying muscle activity and the visual information. By jointly modeling both modalities, the model has been conditioned for generating motion that is consistent with muscle activation.