AGN is a data generation method for skeleton-based human action recognition. AGN can generate diverse new samples for a single or small number of samples, which improves the ability to generalize action recognition.
This repo contains the results of AGN's visualization of seen actions, unseen actions, and more examples. Each visualization image contains the source action, the target action, and the generated action.