Brno University of Technology at TRECVid 2010 SIN, CCD

1. The runs differ in the types of visual features used. All runs use several bag-of-word representations fed to separate linear SVMs and the SVMs were fused by logistic regression. *F_A_Brno_resource_4: Only single best visual features (on the training set) are used – dense image sampling with rgb-SIFT. * F_A_Brno_basic_3: This run uses dense sampling and Harris-Laplace detector in combination with SIFT and rgb-sift descriptors. * F_A_Brno_spacetime_1: This run extends F_A_Brno_color_2 by adding space-time visual features STIP and HESSTIP. 2. Combining multiple types of visual features improves results significantly. F_A_Brno_color_2 achieve more than twice better results than F_A_Brno_resource_4. The space-time visual features did not improve results. 3. Combining multiple types of visual features is important. Linear SVM is inferior to non-linear SVM in the context of semantic indexing. 1. Two runs submitted, but with similar settings; the difference is only in amount of processed test data (40% and 60%) • brno.m.*.l3sl2: SURF, bag-of-words (visual codebook: 2k size, 4 nearest neighbors used in soft-assignment), inverted file index, geometry (homography) based image similarity metric 2. What if any significant differences (in terms of what measures) did you find among the runs? • only one setting used – no differences 3. Based on the results, can you estimate the relative contribution of each component of your system/approach to its effectiveness? • slow search in reference dataset due to unsuitable configuration of used visual codebook 4. Overall, what did you learn about runs/approaches and the research question(s) that motivated them? • change the way of describing the video content – frame based (or key-frame based) approach is not sufficient