Galleries

This research tackles three key challenges in human motion analysis - motion perception anomalies, negative transfer due to limited samples, and catastrophic forgetting in non-stationary data distributions. By leveraging spatiotemporal graph models, domain adaptation, and continual learning, innovative methods were developed to address these issues, achieving significant advancements in aerospace training and medical rehabilitation applications.

This proposal explores innovative solutions for Multi-Modal Action Quality Assessment, focusing on addressing challenges such as modality heterogeneity, missing data, and lack of interpretability in current methods.