Recent advances in emerging technologies such as artificial intelligence and extended reality have pushed the Metaverse, a virtual, shared space, into reality. In Metaverse, users can customize virtual avatars to experience a different life. While impressive, avatar construction requires a lot of data that manifest users in the physical world from various perspectives, and wireless sensing data is one of them. For example, machine learning (ML) and signal processing can help extract information about user behavior from sensing data, thereby facilitating avatar behavior construction in the Metaverse. This article presents a wireless sensing dataset to support the emerging research on Metaverse avatar construction. Rigorously, the existing data collection platforms and datasets are analyzed first. On this basis, we introduce the platform used in this paper, as well as the data collection method and scenario. We observe that the collected sensing data, i.e., channel state information (CSI), suffers from a phase shift problem, which negatively affects the extraction of user information such as behavior and heartbeat and further deteriorates the avatar construction. Therefore, we propose to detect and correct this phase shift by a sliding window and phase compensation, respectively, and then validate the proposed scheme with the collected data. Finally, several research directions related to the avatar construction are given from the perspective of datasets.