MUVS: a Dataset of Multi-View User-Generated Sports Video and Sensor Data

MUVS is a multimodal dataset of multi-view user-generated sports video and sensor data collected under real mobile acquisition conditions. It contains synchronized handheld recordings of live sports events acquired simultaneously from different viewpoints, together with device-level sensor measurements recorded during video capture. The dataset covers 17 events across Basketball, Soccer, Hockey, and Volleyball, comprising 5,039 temporal fragments of 3 seconds each and 258 period-level videos hosted on Google Cloud Storage. Fragment-level camera selection annotations are included. A download script is provided to retrieve the period videos from GCS without credentials.