The workflow for Pixel Classification + Object Classification permits me to select multiple images to process, but the workflow selects one for me during training and I don’t see a way to switch between images to manually classify pixels and/or objects. This is necessary in my case because some of my images have slightly different characteristics, and I want the training to generalize those away.
I ended up using GIMP to manually create this mosaic sampling from all the other images and entering it as the first image so that I could use it to train.
this switching between images is a bit hidden in ilastik is a bit hidden. The selector for the current image is above the layer stack:
On a slightly related note: We usually not recommend to use the combined pixel and object classification workflow. It is there mainly for quick demos. For serious work we would recommend to split the workflow into two steps: first do pixel classification, export the result and use it in a second, object classification project. Memory requirements for the combined workflow can explode quite quickly.