### Table 2: 3D point reconstruction error for the RADIUS image data.

2000

"... In PAGE 24: ... Our algorithm recovered 61 correspondence rooftop polygon points, all of them correct (Figures 11 and 12). The corresponding 3D points reconstructed by the al- gorithm are reported in Table2 . This experiment uses the ground truth 3D data supplied in the model board 1 data set Here we only reported the comparisons between the reconstructed 3D points and their ground truth data for those 3D points whose ground truth coordinates are available.... In PAGE 24: ... This experiment uses the ground truth 3D data supplied in the model board 1 data set Here we only reported the comparisons between the reconstructed 3D points and their ground truth data for those 3D points whose ground truth coordinates are available. From Table2 , we can see that for some image correspondences such as 20, 21, 34, 35, 36, 49, 50, and 57, the trian- gulated 3D points have large errors although their correspondences are determined correctly by our algorithm. This is due mainly to the errors in the locations of rooftop polygon points, since it is well known that these 2D errors have a significant effect on the triangulated 3D data, especially when there are only two images [17].... ..."

Cited by 5

### Table 1 lists the various statistical estimators of the er- rors for all 100 samples. Table 2 demonstrates that our cost value based on the difference in 2D silhouette images has strong correlation with L2 distance in 3D. Also, by comparing all 100 reconstructed 3D faces to the original faces visually, we could see the L2 error has strong cor- relation with the visual similarity of two 3D faces. One important conclusion we can draw from this observation is that silhouette matching with sufficiently large number of viewpoints provides us with a very good estimate of the shape of a human face assuming that the target face is already in the 3D face space that is spanned by the eigen- heads.

2003

"... In PAGE 8: ... Table1 : Statistical estimators of errors... ..."

Cited by 7

### Table 1 lists the various statistical estimators of the er- rors for all 100 samples. Table 2 demonstrates that our cost value based on the difference in 2D silhouette images has strong correlation with L2 distance in 3D. Also, by comparing all 100 reconstructed 3D faces to the original faces visually, we could see the L2 error has strong cor- relation with the visual similarity of two 3D faces. One important conclusion we can draw from this observation is that silhouette matching with sufficiently large number of viewpoints provides us with a very good estimate of the shape of a human face assuming that the target face is already in the 3D face space that is spanned by the eigen- heads.

2003

"... In PAGE 7: ... Table1 : Statistical estimators of errors... ..."

Cited by 7

### Table 1 lists the various statistical estimators of the er- rors for all 100 samples. Table 2 demonstrates that our cost value based on the difference in 2D silhouette images has strong correlation with L2 distance in 3D. Also, by comparing all 100 reconstructed 3D faces to the original faces visually, we could see the L2 error has strong cor- relation with the visual similarity of two 3D faces. One important conclusion we can draw from this observation is that silhouette matching with sufficiently large number of viewpoints provides us with a very good estimate of the shape of a human face assuming that the target face is already in the 3D face space that is spanned by the eigen- heads.

2003

"... In PAGE 6: ... Table1 : Statistical estimators of errors... ..."

Cited by 7

### Table 1: E ect of Bias on 3D Face Reconstruction

"... In PAGE 10: ...quation (9). We present here the results on the rst ve face models in the above mentioned database. Following the convention on the website, we refer to the ve subjects as quot;frame001 quot; to quot;frame005 quot;. From Table1 , we see that the peak value of the bias is a signi cant percentage of the true depth value. This happens only for a few points; however it has signi cant impact on the 3D face model because of interpolation techniques which, invariably, are a part of any method to build 3D models.... In PAGE 10: ... This happens only for a few points; however it has signi cant impact on the 3D face model because of interpolation techniques which, invariably, are a part of any method to build 3D models. The third and fourth columns in Table1 represent the root mean square (RMS) error of the reconstruction represented as a percentage of the true depth and calculated before and after bias compensation. The change in the average error after bias compensation is very small.... ..."

### Table 1. Datasets. We applied our reconstruction algorithm to four datasets. Images is the number of captured images and points is the number of reconstructed 3D points.

2006

"... In PAGE 5: ... Results and Discussion In this section, we present results and comparisons of applying our novel formulation for bundle adjustment to several example scenes. Table1 provides a summary of our three test datasets. Our datasets range from 96 to 32688 reconstructed points and from 48 to 2644 images.... ..."

Cited by 1

### Table 3. Morphological Parameters

"... In PAGE 9: ... (2000), with few outliers. Our adopted WIYN values for Reff, C, and A are listed in Table3 . For all subsequent discussions we will use these values, as determined from the WIYN images.... In PAGE 10: ...distance is listed in Table3 and is used to calculate distance-dependent quantities such as H i mass, luminosity, and linear size. We have also attempted to derive the dynamical masses for these galaxies.... ..."

### Table 1. Comparison of some 3D reconstruction methods. Lexicographical ordering was used so that (i) the importance of a criterion decreases from the first to the last column and (ii) the quality of the method decreases from top to down

2002

"... In PAGE 5: ... Subsequent images are taken one after another and used to extend and improve actual reconstruction. Table1 summarizes the differences among the mentioned methods. Jacobs [8] solves reconstruction with occlusions for orthographic camera, Sturm amp; Triggs [11] solve reconstruction without occlusions for perspective cam- era.... ..."

Cited by 20

### Table 1 Reconstruction error per DOF (in normalized coordinates) Data used 3D reconstruction error

"... In PAGE 12: ...ut the x locations were not. Fig. 7 shows the coordinates used and a sample reconstruction for this experiment. Note that in both cases (see Table1 ), the re- construction is quite accurate in terms of mean- squared error. This shows that the ten learned modes are a su ciently strong characterization to accurately reconstruct the 3D lip shape from 2D data.... ..."

### Table 6 Multiplierless 32d QMF filter banks found by DLM with static and adaptive weightsa

1998

"... In PAGE 23: ...ll the coe cients to be 0.9474. After multiplying each coe cient by this scaling factor and restricting each coe cient to a PO2 form with a maximum of 6 ONE bits, we apply DLM with both static and dynamic weights to find the best PO2 designs. Table6 compares the objectives of the designs found and the corresponding convergence times of DLM. Using the adaptive DLM, all the searches con- verge, and most designs have better reconstruction errors.... ..."