### Table 1. Descriptions of surface representations with the convolution kernels used

2004

Cited by 4

### Table 1. The eight coe cients of the polynomials of the cubic-convolution kernel as a function of the free parameter .

1999

Cited by 7

### Table 3. The 32 coe cients of the polynomials of the septic-convolution kernel as a function of the free parameter .

1999

Cited by 7

### Table 1 Example generalized convolution instantiations Kernel operation C13 c

2002

Cited by 23

### Table 1. Values of the free parameter for the cubic (n = 3), quintic (n = 5), septic (n = 7), and nonic (n = 9) convolution kernels de ned in Section 3.3, resulting from the slope constraint ( amp;), continuity constraint ( ), and flatness constraint ( [), respectively.

2001

Cited by 15

### Table 1: Evaluation of context-sensitive convolution tree kernels using SPT on the major relation types of the ACE RDC 2003 (inside the parentheses) and 2004 (outside the parentheses) corpora.

"... In PAGE 7: ... Context-Sensitive Convolution Tree Kernel In this paper, the m parameter of our context-sensitive convolution tree kernel as shown in Equation (3) indicates the maximal length of root node paths and is optimized to 3 using 5-fold cross validation on the ACE RDC 2003 training data. Table1 compares the impact of different m in context-sensitive convolution tree kernels using the Shortest Path-enclosed Tree (SPT) (as described in Zhang et al (2006)) on the major relation types of the ACE RDC 2003 and 2004 corpora, in details. It also shows that our tree kernel achieves best performance on the test data using SPT with m = 3, which outperforms the one with m = 1 by ~2.... In PAGE 7: ... This may be due to that, although our experimentation on the training data indicates that more than 80% (on average) of subtrees has a root node path longer than 3 (since most of the subtrees are deep from the root node and more than 90% of the parsed trees in the training data are deeper than 6 levels), including a root node path longer than 3 may be vulnerable to the full parsing errors and have negative impact. Table1 also evaluates the impact of entity-related information in our tree kernel by attaching entity type information (e.g.... ..."

### Table 2: PSNRn5bdBn5d results for convolution with the 5 n02 5 uniform averaging kernel.

1999

Cited by 4

### Table 3: Contents of CT Image Database To generate a global texture signature describing an image, we rst calculated texture features for each pixel. For a given database image, we rst convolved it with a number of Laws apos; convolution kernels. We then replaced each pixel value by the sum of the absolute values of the pixel values in a square neighborhood surrounding it: Inew(x; y) = x+N X

1994

"... In PAGE 8: ...We applied CANDID to this problem of retrieving pulmonary CT imagery from a database containing a total of 220 lung images taken from pulmonary CT studies of 34 di erent patients (see Table3 ). Each image was 512 512 pixels in size, consisting of 12-bit grayscale data.... ..."

Cited by 8

### Table 1. Frames per second for various configurations of the algorithm on height fields of size BDBEBK A2 BDBEBK and BEBHBI A2 BEBHBI, using a BJ A2 BJ convolution kernel. HW/No PT: Hardware implementation , pixel texturing disabled. HW/SWPT: Hardware implementation with software pixel texturing. SW: Software implementation. Sizes are in pixels.

"... In PAGE 9: ... 5.1 Frame Rates Table1 reports the frame rates for three configurations: The hardware implementa- tion without pixel textures, which gives incorrect images but has times closest to those likely if pixel texturing were indeed supported, the hardware implementation with soft- ware pixel textures, which provides a lower bound on the frame rate, and the software implementation. The frame rates of Table 1 are those required to generate the raster images of caus- tics from raster images of height fields.... In PAGE 9: ...1 Frame Rates Table 1 reports the frame rates for three configurations: The hardware implementa- tion without pixel textures, which gives incorrect images but has times closest to those likely if pixel texturing were indeed supported, the hardware implementation with soft- ware pixel textures, which provides a lower bound on the frame rate, and the software implementation. The frame rates of Table1 are those required to generate the raster images of caus- tics from raster images of height fields. Running times for height field generation on an Infinite Reality Onyx 2 are 178 frames per second for a BDBEBK A2 BDBEBK heightfield with one impulse (plus four reflected impulses), and 64 frames per second for a BEBHBI A2 BEBHBI heightfield with three impulses (plus 12 reflected impulses).... In PAGE 9: ... Running times for height field generation on an Infinite Reality Onyx 2 are 178 frames per second for a BDBEBK A2 BDBEBK heightfield with one impulse (plus four reflected impulses), and 64 frames per second for a BEBHBI A2 BEBHBI heightfield with three impulses (plus 12 reflected impulses). From Table1 , it is clear that interactive rates of between 9 and 15 frames per second can be achieved on the Infinite Reality for a height field of size BEBHBIA2BEBHBI. On the Indigo 2, the frame rate drops to between three and four frames per second which, while not interactive, is still reasonable.... ..."

### Table V shows that again the stream model provides the best runtime, with the convolution runtimes being about the same as in the per-kernel approach. In the next subsection we compare the approaches and summarize the results.

2007