Image Annotation via Reconstitution Graph Learning Model

Table 1

Details of the training set of the two datasets. The number of images and labels are given in the format mean/maximum.

Dataset	Corel 5K	IAPR TC12
of img.	4999	19627
Vocab. size	260	291
Training img.	4500	17665
Testing img.	499	1962
Labels per img.	3.4/5	5.836/10.04
Img. per label	5.7/23	347.7/4999