SemOI: A Semantical Augmentation for the Open Images Detection

Abstract

In this paper, we present a semantic extension to the popular Open Image dataset. Using a python script, we transferred the JSON-representation into RDF and then extended the 576 detectable classes by 392 scenes connected by 57 intermediate objects called occasions. The extension allows the inferring of possible locations and semantic suitable objects for the detected items. Using a GUI, a user can upload a picture and calculate a numerical value for the possibility of a given scene.

Keywords

• open image
• semantic enhancement
• scene detection