Table 1.
AP scores on VOC2009 test data with fixed -norm.
Table 2.
AP scores obtained on the VOC2009 training data set with fixed -norm.
Table 3.
Average AP scores obtained on the ImageCLEF2010 test data set with -norm fixed for all classes.
Table 4.
Average AP scores on the VOC2009 test data with -norm class-wise optimized on training data.
Table 5.
Average AP scores on the ImageCLEF2010 test data with -norm class-wise optimized.
Figure 1.
Similarity of the kernels for the VOC2009 (Top) and ImageCLEF2010 (Bottom) data sets in terms of pairwise kernel alignments (Left) and kernel target alignments (Right), respectively.
In both data sets, five groups can be identified: ‘BoW-S’ (Kernels 1–15), ‘BoW-C’ (Kernels 16–23), ‘products of HoG and HoC kernels’ (Kernels 24–27), ‘HoC single’ (Kernels 28–30), and ‘HoG single’ (Kernels 31–32).
Figure 2.
Histogram of kernel weights as output by
-norm MKL for the various classes on the VOC2009 data set (32 kernels×20 classes, resulting in 640 values).
-norm (top left)),
-norm (top right),
-norm (bottom left), and
-norm (bottom right).
Figure 3.
Images of typical highly ranked bottle images and kernel weights from
-MKL (left) and
-MKL (right).
Figure 4.
Images of a typical highly ranked cow image and kernel weights from
-MKL (left) and
-MKL (right).
Table 6.
AP Scores and standard deviations showing amount of randomness in feature extraction: Results from repeated computations of BoW Kernels with randomly initialized codebooks.
Table 7.
MKL versus Prior Knowledge: AP Scores with a smaller fraction of well scoring kernels.
Figure 5.
Diversity measure between correctly classified samples for all pairs of 32 kernels.
Left: Average over all concept classes. Right: Maximum over all concept classes.
Table 8.
AP Scores in Toy experiment using Kernels with disjoint informative subsets of Data.