I'm new to GeoNet and hopefully this is an appropriate place to ask this! I'm working with the deep learning tools and exported some data to use for a PSPNet model. However, after looking at the stats file in the exported data from ArcGIS, the classes are very unbalanced. I'm looking to under sample images from the heavier classes to pair them down to match the constraining class so my dataset is balanced. Is there a tool in ArcGIS for this? Or is there a good way to undersample the exported training data outside of ArcGIS?
There is this "class_balancing" parameter which you can set to True in case of Segmentation models in ArcGIS API for Python. See if you can make use of that.