Semisupervised Vector Quantization in Visual SLAM Using HGCN

Table 1

Exploring spatial models: unrestricted model with unique location values, FABMAP2’s log-likelihood consistency for unobserved “q” values, and the normalized interpretation of FABMAP2.

a
b
c		0	0

Part (a) depicts an unrestricted model where each location can take a specific value. In part (b), the FABMAP2 model is shown in which log-likelihood of locations where “q” values are not observed must take the same value. Part (c) shows the normalized version of part (b) [1].