Human action/pose synthesis and prediction




Collecting and annotating human action datasets is expensive. Can we automate this process by 3D graphics engines? This project aims to synthesize various human actions and poses and use this dataset to train robust action/pose recognition models that are of value in real-world applications. Perspective candidates are expected to have experience in computer vision research and have strong coding and English abilities.


Updated:  10 August 2021/Responsible Officer:  Dean, CECS/Page Contact:  CECS Marketing