Overview
StarDist is a Python-based tool for object detection, particularly focused on segmenting objects with star-convex shapes in 2D and 3D images. It utilizes a deep learning model trained to predict distances from each pixel to the object boundary along a fixed set of rays, as well as object probabilities. The core architecture involves training a convolutional neural network on pairs of raw images and corresponding labeled images, where each pixel is assigned a unique object ID. The final segmentation is achieved through non-maximum suppression (NMS) of candidate polygons/polyhedra generated from the model's predictions. Key use cases include cell nuclei segmentation in microscopy images, histopathology image analysis, and general object detection in biological imaging applications. The tool is designed for researchers and practitioners in the fields of biology, medicine, and computer vision.
