Keywords: cell analysis, instance segmentation, object detection
TL;DR: In this paper, we proposed a instance segmentation method based on a object detection model and a semantic segmentation model.Our method reduce 30% inference on larger images.
Abstract: ell instance segmentation, which identifies each specific cell area within a microscope image, is useful for cell analysis. Because of the high computational cost brought on by the large number of objects in the scene, mainstream instance segmentation techniques require much time and computational resources. In this paper, we proposed a two-stage method in which the first stage is detecting the bounding boxes of cells, and the second stage is segmentation in the detected bounding boxes. This method reduces inference time by more than 30% on images that image size is larger than 1024 pixels by 1024 pixels compared to the mainstream instance segmentation method while maintaining reasonable accuracy without using any external data.