Abstract

Fine-grained action recognition datasets exhibit environmental bias, where multiple video sequences are captured from a limited number of environments. Multi-modal nature of video（视频的多模态性），提出的方法一个是multi-modal self-supervision，还有一个是adversarial training per modality

Introduction

fine-grained action recognition,


Few works have attempted deep UDA for video data《Temporal attentive alignment for large-scale video domain adaptation, ICCV2019》《Deep domain adaptation in action space, BMVC2018》

Conclusion

modality指的是两种信息（optical flow和RGB信息）,future work包含audio

Key points: Motivation很好; 提出的新数据集