A Review on Instance Segmentation Using Mask R-CNN

Abstract

Instance segmentation consolidates object detection, where the objective is to classify and localize each objects using a bounding box, and semantic segmentation, where the objective is to characterize every pixel into the given object classes. Mask R-CNN is a deep learning architecture used for instance segmentation. It is an augmentation of the well known Faster R-CNN object detection architecture. Mask R-CNN adds an additional mask branch to the existing Faster R-CNN model. The Faster R-CNN produces two things for each object in the picture. Its class label and the bounding box co-ordinates. Mask R-CNN adds an extra branch to this which yields the object mask too. Mask prediction is done in corresponding with bounding box creation and grouping. This paper contains the idea of how Mask R-CNN performs instance segmentation by using examples of vehicle damage detection and segmentation, Detection and segmentation of oral diseases and segmentation of news paper elements.