한보형 (서울대 / 부교수)
Real-Time Visual Tracking by Convolutional Neural Networks
Recent tracking algorithms based on convolutional neural networks achieve impressive accuracy, and MDNet is one of the seminal works among them. However, there are still many remaining issues, especially in terms of efficiency. I present a novel real-time visual tracking algorithm based on MDNet, where feature extraction time is significantly reduced by RoIAlign operations and a novel discriminative instance embedding loss is integrated in the pretraining stage to improve accuracy. The proposed real-time MDNet runs at approximately 50 frames/second with almost same accuracy with MDNet. Contrary to many existing approaches, our algorithm works very well on various datasets without dataset sensitive parameter tuning.
Bohyung Han is currently an Associate Professor in the Department of Electrical and Computer Engineering at Seoul National University, Korea. Prior to the current position, he was an Associate Professor in the Department of Computer Science and Engineering at POSTECH, Korea and a visiting research scientist in Machine Intelligence Group at Google, Venice, CA, USA. He received the B.S. and M.S. degrees from Seoul National University, Korea, in 1997 and 2000, respectively, and the Ph.D. in Computer Science at the University of Maryland, College Park, MD, USA, in 2005. He served or will be serving as an Area Chair or Senior Program Committee member of major conferences in computer vision and machine learning including CVPR, ICCV, NIPS, IJCAI and ACCV, a Tutorial Chair in ICCV 2019, and a Demo Chair in ACCV 2014. His research group won Visual Object Tracking (VOT) Challenge in 2015 and 2016.