Computer Vision News - February 2023

31 SpectrAI Installing SpectrAI is super easy. It requires Python 3.8 so make sure you have it installed on your machine. We used this line on terminal (for Mac) to set it up as default python3: ln -s -f /usr/local/bin/python3.8 /usr/local/bin/python3 We can install it through pip: pip3 install spectrai Or cloning the github repository. I suggest this option as more complete. git clone git@github.com :conor-horgan/spectrai.git cd spectrai pip3 install -e . Once the toolkit is successfully installed, to train a network, one can run the command: spectrai_train . This calls the python script train.py which automatically sets pre-processing and network parameters, by reading a config file. By default, the config file called is one for multi-resolution task. For a new task, you can choose between several available networks (DenseNet, EfficientNet, RCAN, ResNet, SegNet, U-Net) and standard augmentation types (e.g. flip, crop) or spectral specific ones (e.g. spectral shift, spectral flip). Depending on your application, you can choose the desired combination and create your own config file, which can be executed by typing the following command: spectrai _ train --config custom _ config.yml –verbose See a nice image of spectral image lambda stack here.