PMC

Influence of silhouette reprojection error in the second phase optimization. (a) is the reconstruction result of dino sparse ring by including the silhouette reprojection error in the energy function. (b) is the reconstruction result of dino sparse ring without using the term of silhouette reprojection error by setting λ to 0. (c) One of silhouette images in 16 viewpoints. (d) Predicted silhouette image generated via surface (a). (e) Predicted silhouette image generated via the surface (b). (f) The inconsistent regions of (d) and (c). (g) The inconsistent regions of (e) and (c).

Example of Omnimatte [] silhouette estimation performance with voxel projections onto the image (data sample #9): (a) input RGB image; (b) silhouette image with magnified silhouette quality values on its edge; (c) voxel center projections onto the silhouette. The projections in the image are marked with pixel-sized dots. The blue pixels represent the silhouette, and the red represent the background.

Details of the Silhouette scoring. (A) Silhouette plot offering a graphical representation of consistency within the three clusters (C1 in blue, C2 in green and C3 in magenta) identified by k-means with k = 3. The Silhouette coefficient values, reported on each bar of the plot, represent a measure of how similar each data instance (i.e., each RSN) is to its own cluster in comparison to other clusters. Specifically, a silhouette coefficient close to 1 indicates that the RSN is close to the center of the cluster, while RNSs with silhouette coefficients close to 0 are on the border between two neighboring clusters. The average value of the Silhouette coefficients reported on the bars is 0.652 and represents the final Silhouette score for k = 3. (B) Plot showing the average Silhouette scores for different values of k (i.e., different number of k clusters), with 2 ≤ k ≤ 10. Note that, for our data, the maximum Silhouette score (0.652), which corresponds to the optimal number of k clusters to be used for the analysis, is obtained for k = 3 (Silhouette score = 0.652).

Silhouette analysis revealed that the optimal number of clusters for the Semantic model was 6. The analysis was performed using the fviz_nbclust function of the R package ‘factoextra’. The silhouette width is a measure of how well each item fits within its cluster with higher values indicating better fit. The average silhouette width is the average of the all items. The algorithm computes the average silhouette width as a function of the number of clusters (k). The highest value of the average silhouette width is taken and the corresponding k is selected as the optimal number of clusters.