20160615-19蓝振忠:Towards usable multimedia event detection (MED) from web vide ...

报告嘉宾2：蓝振忠（卡内基梅隆大学（CMU））

报告时间：2016年6月15日（星期三）晚21:00（北京时间）

报告题目：Towards usable multimedia event detection (MED) from web videos

主持人：  高陈强（重庆邮电大学）

报告摘要：Multimedia data have proliferated in the past few years, ranging from ever-growing personal video collections to films and professional documentary archives. Numerous tools and applications have been invented to describe, organize, and manage video data. Previous research mainly focuses on recognizing scene, object and action. However, these atomic concepts are too primitive to be used for users to search videos from data collections. We instead focus on detecting higher level event in unconstrained videos. In this talk, I will present several techniques that significantly contribute to the leading performance CMU team in TRECIVD MED competition 2011 ~ 2015. As we mostly focus on improving motion representations for unconstrained videos, these techniques are also useful for action recognition.

参考文献：

[1] Lan, Z., Lin, M., Li, X., Hauptmann, A. G., & Raj, B. (2015). Beyond Gaussian pyramid: Multi-skip feature stacking for action recognition. In CVPR'15. 

[2] Lan, Z. Z., Bao, L., Yu, S. I., Liu, W., & Hauptmann, A. G. (2014). Multimedia classification and event detection using double fusion. Multimedia Tools and Applications, 71(1), 333-347.

[3] Lan, Z., Yu, S. I., & Hauptmann, A. G. (2015). Improving Human Activity Recognition Through Ranking and Re-ranking. arXiv preprint arXiv:1512.03740.

报告人简介：

Zhenzhong Lan recevied his bachelor degree of software engineering and statistics from Sun Yat-sen University in 2010. He is currently a 4th year PhD student working with Dr. Alexander G. Hauptmann in Informedia Group, CMU, Pittsburgh, PA, USA. His principal interests lie in the development of computer vision and multimedia techniques for event and action recognition in large scale video. His has more than 20 publications that have received over 400 citations.