VIREO @ TRECVID 2012: Searching with Topology, Recounting will Small Concepts, Learning with Free Examples

Wei Zhang, Chun Chet Tan, Shiai Zhu, Ting Yao, Lei Pang, Chong-Wah Ngo
2012 TREC Video Retrieval Evaluation  
The vireo group participated in four tasks: instance search, multimedia event recounting, multimedia event detection, and semantic indexing. In this paper, we will present our approaches and discuss the evaluation results. Instance Search (INS): We submitted four Bag-of-Words (BoW) based runs this year to mainly test the proper way of exploiting spatial information through comparing the weak consistency checking (WGC) and our spatial topology consistency checking using Delaunay Triangulation
more » ... ) based matching. Considering the special features of the INS task of TRECVID (e.g., multiple image examples for a query; ROI indicating the spatial location of the instance), we also study the effects of multi-query fusion and background context modeling on top of BoW retrieval system. -F X NO vireo bl 4: Baseline run with standard weak geometric consistency checking (WGC [1]), background context modeling, and video level fusion. -F X NO vireo dtcv 3: Spatial topology consistency checking via Delaunay Triangulation (DT). This run is similar with vireo bl, except we use DT instead of WGC for spatial checking. -F X NO vireo dtc 2: Spatial run with DT and background context modeling. Compared with vireo dtcv, we do not use video level fusion for this run. -F X NO vireo dto 1: Spatial matching with DT by using only the ROI region containing the object.
dblp:conf/trecvid/0031TZYPN12 fatcat:5vplqjsezvhmhl7tcu3h2pwipm