U-Net

Overview

U-Net is a convolutional network architecture designed for biomedical image segmentation. It excels in scenarios requiring fast and precise segmentation, outperforming previous methods on challenges like neuronal structure segmentation in electron microscopic stacks. Its architecture features a contracting path (left side) to capture context and a symmetric expanding path (right side) that enables precise localization. Skip connections pass feature maps from the contracting path to the expanding path, preserving high-resolution information. The network is trained end-to-end from very few images and relies on heavy data augmentation to use the available annotated samples more efficiently. U-Net has achieved state-of-the-art results in various ISBI challenges, including cell tracking and caries detection in bitewing radiography. The provided release includes the trained network, source code, and necessary libraries for deployment on Ubuntu Linux 14.04 with Matlab 2014b (x64).

Common tasks

Image Segmentation Object Detection Image Analysis

FAQ

View all

What is U-Net?

U-Net is a convolutional network architecture designed for fast and precise segmentation of images, particularly in biomedical applications.

What are the system requirements for running U-Net?

U-Net is primarily tested and compiled on Ubuntu Linux 14.04 and Matlab 2014b (x64). It can benefit from a CUDA-capable GPU.

Where can I download the U-Net release?

The U-Net release can be downloaded from the provided link on the website in the form of a .tar.gz archive.

How do I apply segmentation and tracking to my images?

After downloading and extracting the release, you can use the provided shell script (segmentAndTrack.sh) to apply the segmentation and tracking algorithms. Ensure you adjust GPU settings if necessary.

FAQ+