Hierarchy-NMS: Merging Candidate Bounding Boxes for Cerebrospinal Fluid Cell Image Segmentation
Xianwei Xu, Fangqi Li, Shilin Wang, Zhenhai Wang · Journal of Physics Conference Series · 2020
Abstract In this article, Hierarchy-Non Maximum Suppression (H-NMS) is proposed to fix the defects of general NMS algorithms in cerebrospinal fluid cell image segmentation. To cope with the confusion caused by the hierarchy of candidate bounding boxes in this scenario, a tree structure is built on the top of all bounding boxes and only leaf nodes are saved as the final output. The experimental results showed that H-NMS outperforms some variants of NMS in the cerebrospinal fluid cell image processing in both image segmentation and cell counting.