MMAct Multimodal Action Understanding Dataset

MMAct stands for Multi Modal Action, which is a large-scale dataset for multimodal/cross-modal action understanding. The dataset was recorded by 20 subjects in 7 modes, namely RGB video, key points, acceleration, gyroscope, direction, Wi-Fi, and pressure signals. The dataset includes more than 36,000 video clips of 37 types of actions, covering a wide range of daily activities.