期刊名称：International Journal on Smart Sensing and Intelligent Systems
印刷版ISSN：1178-5608
出版年度：2015
卷号：8
期号：2
页码：1284-1312
出版社：Massey University
摘要：The quality of a semantic annotation is typically measured with its averagedclass-accuracy value, whose computation requires scarce ground-truth annotations.We observe that humans accumulate knowledge through their vision and believethat the quality of a semantic annotation is proportionally related to its compatibilitywith the vision-based knowledge. We propose a knowledge-compatibility benchmarker,whose backbone is a regression machine. It takes as input a semantic annotation andthe vision-based knowledge, then outputs an estimate of the corresponding averagedclass-accuracy value. The knowledge encodes three kinds of information, namely: cooccurrencestatistics, scene properties and relative positions. We introduce three typesof feature vectors for regression. Each specifies the characteristics of a probability vectorthat captures the compatibility between an annotation and each kind of the knowledge.Experiment results show that the Gradient Boosting regression outperforms then-Support Vector regression. It achieves best performance at an R2-score of 0.737 andan MSE of 0.034. This indicates not only that the vision-based knowledge resembleshumans’ common sense but also that the feature vector for regression is justifiable.
关键词：vision-based knowledge; knowledge-compatibility benchmarker; semantic segmentation;averaged class accuracy; regression
