A Simultaneous Object Detection and Component Segmentation Approach Based on Mask R-CNN
Deshuai Zheng,Jin Yan,Yong Liu,Tao Xue
DOI: https://doi.org/10.1109/cyber50695.2020.9279114
2020-01-01
Abstract:Instance segmentation is considered to be one of the critical tasks in image understanding. Recently, Mask R-CNN has established itself as a simple, flexible and effective model in image instance segmentation by demonstrating outstanding performance. However, under specific situations, we not only need to segment the objects, but also need its precise components information and existing instance segmentation architectures including Mask R-CNN are unable to achieve this task. In this paper, we propose a simultaneous object detection and component segmentation approach based on Mask R-CNN. Our architecture contains two main branches, in which the object path is to output the object bounding boxes and the component path is to obtain the component bounding boxes as well as the component mask. For the overlapping problem, we defined an Overlapping Area Ratio(OAR) between object bounding box and component bounding box to confirm their relationships. The result shows that our approach is a practicable way for object and component segmentation, and it's the beginning of object detection and component segmentation.