每日（一文 Multi-Domain and Multi-Task Learning for Human Action Recognition

What to Focus on?

Though this ariticle mainly focuses on human activity recognition with video datasets, i.e. the conventional and representitve deep learning task, it doesn’t mean that there is nothing valuable for radar research. The human activity recognition task for radar datasets actually shares the multi-view problem in common with the video counterpart.

In the former article [ref: Omnidirectinal], this problem is basically solved by proposing a novel architecture and expanding the variety of the view-angles of the simulation dataset, which, though, does not answer why the proposed architecture/method fit the expanded dataset, i.e. it works because of whether the structure or merely the expansion of the dataset, even if the result really outperforms.

Thus in this article, according to their description, the domain/modal