Semisupervised Vector Quantization in Visual SLAM Using HGCN

Table 2

An overview of the datasets utilized in various studies, including HGCN-FABMAP, HGCN-BoW, HGCN-ORB College, Lip6indoor, and Frieburg3.

Datasets	Number of images		Long range	Dataset description	Ref.
	Train	Test
Lip6indoor dataset	350			Images are taken from a lab environment with a narrow corridor	[23]
TUM sequence-11	1500			Images are taken from a lab environment with different lighting conditions. These data are our training dataset, where Lip6indoor is the test dataset	[24]
New College	1073		●	This dataset has been taken from the Oxford university campus, which includes complex repetitive structures. This is a stereo dataset with left and right sequences. We have used the left sequence containing images with 640 × 480 resolution as our training data for the Newer College dataset	[25]
Newer College	200			This is a large video with the same context of the New College dataset, and it contains 3 loops. The camera is experiencing cluttered movements	[26]
St. Lucia suburbs	540	1000	●	Images are taken by a camera-equipped car. The dataset has two parts: a training dataset and a set of test images as test data	[27]
Freiburg-3	2500			Images taken from the surrounding area of a table and it contains one loop	[28]