nnInteractive : Redefining 3D Promptable Segmentation скачать в хорошем качестве

nnInteractive : Redefining 3D Promptable Segmentation

présenté par : Fabian Isensee and Maximilian Rokuss Helmholtz Imaging - Heidelberg. nnInteractive: Redefining 3D Promptable Segmentation nnInteractive is designed to make 3D segmentation more accurate and efficient across clinical and research imaging. Unlike 2D-based models such as SAM, it is built specifically for 3D data. The method supports different types of user inputs—points, scribbles, boxes, and lassos—to guide the segmentation process interactively. Trained on more than 120 volumetric datasets from various imaging modalities (CT, MRI, PET, 3D microscopy, etc.), nnInteractive provides high accuracy and generalization. It is integrated into common imaging platforms like Napari and MITK Workbench, ensuring easy use in real-world applications. Github : https://github.com/MIC-DKFZ/nnInterac... Manuscript: https://arxiv.org/abs/2503.08373