seabet betting time:Monday, September 16, 2019, 2:00 pm
seabet betting fixed point:Conference Room 409, Computer seabet betting
seabet betting title: Skeleton extraction in natural images
Reporter:Ke Wei Carnegie Mellon seabet betting Robotics Institute
seabet betting Summary:Symmetry is ubiquitous in some objects in nature，Skeleton as a description of symmetry，Is one of the important research topics in computer vision，It can be applied to target characterization、Image foreground extraction, etc.。The skeleton extracted data develops natural images from binary images，The method of skeleton extraction has also developed from traditional digital image processing methods to methods based on deep learning。In this seabet betting，Will introduce Sym-PASCAL, a natural image data set，And side output residual network and its improved method，Further promote research on natural image skeleton extraction。SymPASCAL data set，With multi-target co-occurrence、Occlusion、Difficulties such as complex backgrounds。The side output residual network performs side output estimation at each convolution stage of the fully convolutional network，And stack the residual module on the side output，Driving the residual between the network output and the real skeleton to zero。This method is simple and effective，Achieved excellent performance on Sym-PASCAL and other existing skeleton extraction data sets。Based on the side output residual network，Proposed linear expansion network，Theoretically analyzed it to find the optimal side output residual network structure，Further improved skeleton extraction performance。
About the speaker:Ke Wei，Doctor，Currently a postdoctoral researcher at Carnegie Mellon University’s Robotics Institute。Before this，He received his PhD from the University of Chinese Academy of Sciences in 2018，Received undergraduate degree from Beijing University of Aeronautics and Astronautics in 2011。Received funding from the China Scholarship Council from 2015 to 2016 for joint training at the University of Oulu, Finland。His seabet betting interests include visual object perception and intelligent video analysis。18 papers published so far，includes IEEE CVPR, 5 papers from top computer vision conferences such as ECCV，And serves as a reviewer for many well-known journals and conferences。
Postdoctoral period，He participates in the Deep Intermodal Video Project (DIVA) of the U.S. Intelligence Advanced Research Projects Agency (IARPA)，Conducted in-depth research on the application of behavior recognition in traffic monitoring scenarios，Accumulated rich engineering experience。PhD period，He has made a series of progress in natural image skeleton extraction，The first work was accepted as an oral seabet betting by CVPR 2017，This paper released a new data set and proposed a simple and effective way of side output residual network。The contralateral output residual network will be explained and extended later，Proposed linear expansion network and orthogonal decomposition network，Accepted by ECCV 2018 and CVPR 2019 respectively。Meanwhile，Participate in the skeleton detection competition of ICCV 2017 and CVPR 2019，Won two first prizes respectively，Two second prizes。
