this is currently not directly possible I’m afraid.
But maybe just to clarify, you have (two or more) different types of objects in your image and only want to track a particular type? In that case you could achieve your goal with one more step: object classification. You could train object classification to classify your two (or more) types of objects, export the “Object Probabilities”, which is a multi-channel image with one channel for each class. Each object will be associated with the probability to belong to a specific class. With this image you could go into the tracking workflow with a probability map and threshold the channel you are interested in with a threshold of 0.5.