Software Tools on Qim Center

InSegT3D

Mon, 01 Jan 0001 00:00:00 +0000

InSegT3D is an interactive segmentation tool that utilizes the U-Net deep learning architecture to quickly and efficiently segment 3D volumetric images. By providing a few scribbles, you get a complete segmentation of your 3D dataset. It utilizes the Zarr storage format to enable the segmentation of extremely large images.

Local Thickness

Mon, 01 Jan 0001 00:00:00 +0000

Local thickness is a fundamental morphological measure defined as the radius of the largest sphere that fits inside an object at any given point. Our fast algorithm computes local thickness in a fraction of the time compared to conventional approaches.

Same-Class Neighbor Penalization

Mon, 01 Jan 0001 00:00:00 +0000

Same-Class Neighbor Penalization (SCNP) is a novel optimization method designed to improve the topological accuracy of image segmentation models. SCNP discourages topological errors, such as broken connections, or isolated holes or islands, by penalizing the poorest-classified neighbor of each pixel’s logit. This is achieved through simple min- and max-pooling operations over local neighborhoods, which amplify the loss contribution of pixels that are most likely to create incorrect connectivity.

Structure Tensor

Mon, 01 Jan 0001 00:00:00 +0000

The structure tensor is a 3×3 symmetric positive semi-definite matrix that summarizes orientation in a small neighbourhood around every point in a 3D volume. This tool provides an efficient implementation for computing the structure tensor and extracting dominant orientations and shape measures from large 3D datasets.