学术报告——“Large-Scale Online Near Duplicate Video Detection”
题目：Large-Scale Online Near Duplicate Video Detection
报告人：Wenbo He (Associate Professor, Mcmaster University)
报告内容：Nowadays, billions of videos are hosted and shared in the cloud and there exist a nontrivial portion of the near-duplicate video copies. An efficient and accurate content-based online near-duplicate video detection (NDVD) for large scale data set is a fundamental research which will benefit applications such as duplication-aware storage, pirate video detection, polluted video tag detection, and searching result diversification, etc. In this talk, I will report our progress on NDVD, and the talk contains three parts: (1) I will introduce a system called CompoundEyes, which shifts the focus from feature design or algorithm design to system design, so that both efficiency and accuracy can be achieved for NDVD. (2) I will show the streaming data processing pipeline, called Marlin, to handle online feature extraction for large scale NDVD, and achieve a fine-grained resource allocation among the videos with various resource demands. (3) I will present a practical and highly efficient locality sensitive hashing (LSH) system to provide the large scale indexing structure for streaming data, and deal with concurrent query/update requests.