Frequency based Semantic Consistency Groups in Object Detection

This paper presents the results of applying semantic knowledge to object detection using deep learning. We discover semantic relationships between groups of objects by counting co-occurrence frequencies of object groups before combining them with detection output from Faster R-CNN network in a re-optimization process for better detection. The research was conducted on two datasets – MS COCO and a small self-made dataset consisting of images of beverage items on shelves in several supermarkets.