CAUVIS-IME-USP at TRECVID 2011: instance search

Structured Abstract In this paper, we describe the participation and results of CAUVIS-IME-USP at TRECVID 2011 in instance search pilot task. Briefly, what approach or combination of approaches did you test in each of your submitted runs?  CAUVIS_USP: PHOW descriptor and the bag-of-word approach. What if any significant differences (in terms of what measures) did you find among runs? We submitted just one run. Based on the results, can you estimate the relative contribution of each component of your system/approach to its effectiveness? The best performance of our run was in the two queries of type location that we obtained score above the median of runs. In both locations, the texture of the scenes is the most important characteristic. One of the descriptors we are using is PHOW and it is a very suitable to describe texture. Other important factor is that for both queries the mask are big, including almost all the image. Our method did not work well when the interested instance is small. Overall, what did you learn about runs/approaches and research question(s) that motivated them? PHOW descriptor and the bag-of-word approach is the state-of-art in object recognition but it did not work well in instance search task of TRECVID.