Action Recognition-Temporal Attentive Alignment for Large-Scale Video Domain Adaptation——ICCV2019

Abstract

image-based domain adaptation, domain shift in videos,
Two large-scale DA datasets (UCF-HMDB_full, Kinetics-Gameplay)

Introduction

videos can suffer from domain discrepancy along both the spatial and temporal directions.
Claim的两个点:

Conclusion

the presented datasets is including both real and virtual domains;
Align and learn temporal dynamics；future work, different cross-domain video tasks, open-set setting;

Key points: 代码和数据集开源；写的逻辑更清楚；打的点就是解决real-world problems; video classification和action recognition的区别和关系（相似）; video domain adaptation的相关工作比较少；