In this article, we propose a lightweight and flexible enhanced Tai Chi training system composed of multiple standalone virtual reality (VR) devices. The system aims to enable a hyper-realistic multi-user action training platform at low cost by displaying real-time action guidance trajectories, providing real-world impossible visual effects and functions, and rapidly enhancing movement precision and communication interest for learners. We objectively evaluate participants' action quality at different levels of immersion, including traditional coach guidance (TCG), VR, and mixed reality (MR), along with subjective measures like motion sickness, quality of interaction, social meaning, presence/immersion to comprehensively explore the system's feasibility. The results indicate VR performs the best in training accuracy, but MR provides superior social experience and relatively high accuracy. Unlike TCG, MR offers hyper-realistic hand movement trajectories and Tai Chi social references. Compared with VR, MR provides more realistic avatar companions and a safer environment. In summary, MR balances accuracy and social experience.

We propose a neuralized undirected graphical model called Neural-Hidden-CRF to solve the weakly-supervised sequence labeling problem. Under the umbrella of undirected graphical theory, the proposed Neural-Hidden-CRF embedded with a hidden CRF layer models the variables of word sequence, latent ground truth sequence, and weak label sequence with the global perspective that undirected graphical models particularly enjoy. In Neural-Hidden-CRF, we can capitalize on the powerful language model BERT or other deep models to provide rich contextual semantic knowledge to the latent ground truth sequence, and use the hidden CRF layer to capture the internal label dependencies. Neural-Hidden-CRF is conceptually simple and empirically powerful. It obtains new state-of-the-art results on one crowdsourcing benchmark and three weak-supervision benchmarks, including outperforming the recent advanced model CHMM by 2.80 F1 points and 2.23 F1 points in average generalization and inference performance, respectively.