Computer Vision News - January 2021

This means mainly goodbye long hours spent optimising parameters . Anybody who is trying to implement a good segmentation method, or any type of deep learning model would know how tedious the “refining part” can be. It can take more to adapt a working architecture to a new set of data than to start from scratch. This tool aims to solve this issue. Let’s see how! Start with nnU-Net First of all, it’s ideal to create a new virtual environment. This can be done through either conda or pip. We’ll go for the first option and, after opening a new terminal window, we can type the following commands: Now we have two options: 1) to install nnU-Net through the standard pip installation for use as an out-of-the-box segmentation algorithm, 2) use nnU-Net as an integrative framework, by running: This will install the nnU-Net folder from the github source, hence making all the code accessible and ready for future modifications. 15 nnU-Net conda create -n nnUnet_env python pytorch conda activate nnUnet_env git clone https://github.com/MIC-DKFZ/nnUNet.git cd nnUNet pip install -e