Video Search Engine vitrivr of the Research Group DBIS Winner of the DDPS Competition 2019
The Video Browser Showdown (VBS http://www.videobrowsershowdown.org/) is an annual live video search competition in which international researchers evaluate and demonstrate the efficiency of their explorative video search engines on a common dataset in front of an audience. In a collection of more than 1,000 hours of video material (> 1.3 TB), three different types of queries must be resolved as quickly as possible: i.) Visual known-item search (HIS): a video sequence of 20 seconds is selected from the data set and played back. The participants must find exactly this sequence. ii.) Textual known-item search: the searched video sequence is only described by a text. iii.) Ad-hoc video search (AVS): here a general description is presented (e.g. "Find all shots showing cars in front of trees"); the task is to find as many suitable sequences as possible within the given time. The competition takes place in two stages: in the first phase the video search engines are operated by the respective developers. In the second phase (novice search), randomly selected voluntary participants of the conference have to answer these questions in order to test the user-friendliness of the systems.
The DDPS Competition 2019 took place as part of the MultiMedia Modeling Conference (MMM) in Thessaloniki, Greece. After 2017 Luca Rossetto, Mahnaz Amiri Parian, Ralph Gasser, Ivan Giangreco, Silvan Heller and Heiko Schuldt from the Databases and Information Systems Research Group won the DDPS competition for the second time with their vitrivr (https://vitrivr.org/) video search engine and prevailed against international competition.