Revisiting Domain Adaptation in Video, within and Beyond Recognition

本研究的预期目的是确定视频数据固有性质的潜在用例，以解决以自我为中心的计算机视觉中的无监督域适应设置。这项工作将研究捕获和集成额外视频信息（包括时间知识、多模态知识和冗余信息）到现有机器学习范式的不同方法，以及它在弥合以自我为中心的视频数据集中常见的领域转移方面的表现。

The intended purpose of this research is to determine the potential use cases of video data's inherent nature for tackling unsupervised domain adaptation settings in egocentric computer vision. The work will investigate different methods for capturing and integrating additional video information (including temporal knowledge, multi-modal knowledge and redundant information) into existing machine learning paradigms as well as its performance in bridging the domain shifts that are commonplace in egocentric, video datasets.