Abstract:

https://meet.google.com/jak-swcu-xbv

Video has become one of  the major media in our society, bringing considerable interests in the development of video analysis techniques for various applica tions. Temporal Dynamic, which represents how information changes along time, is the key component for videos. However, it is still not clear how temporal dynamics benefit video tasks, especially for the cross-domain case, which is close to real-world scenarios. In this talk, I will share my Ph.D. work which effectively exploits temporal dynamics from videos to tackle distributional discrepancy problems for video understanding.  For cross-domain action recognition, I collected two large-scale datasets: UCF HMDB and Kinetics-Gameplay to facilitate cross-domain video research, and proposed Temporal Attentive Adversarial Adaptation Network (TA3N) to simultaneously attend, align and learn temporal dynamics across domains. For cross-domain action segmentation, I p roposed Self-Supervised Temporal Domain Adaptation (SSTDA) to jointly align cross domain feature spaces embedded with local and global temporal dynamics.

Bio:

Min Hung Chen is currently a Senior AI Engineer at MediaTek Inc. He obtained the Ph.D. degree from Georgia Institute of Technology in 2020. He received his Bachelor's and Master's degrees in 2010 and 2012, respectively, both from National Taiwan University. Before studying for Ph.D., Min Hung was a research assistant at Academia Sinica, working on Multi-modal Action Recognition under Prof. Yen Yu Lin's supervision. Min Hung 's main research interest is Video Understanding beyond Fully Supervision, including domain adaptation, transfer learning, action recognition, etc. During his Ph.D., he has publishe d several papers at top conferences, including CVPR and ICCV, and interned at several companies like Aipoly, PlayStation, and Baidu USA. He also served as the reviewer in many top conferences, such as CVPR, ICCV, ECCV, ICML, NeurIPS. Recently, he co-organi zed the Mobile AI Workshop with ETHZ at CVPR 2021.