Assuming my typical training data are images that each contain multiple animals, with the animals themselves already located/segmented (We have information for each pixel whether it is an animal or not). The label data is information on how these animals are grouped into clusters. For instance, a picture can contain three clusters of animals.
How do we train a model to cluster these animals into groups, using the whole initial image, information from the Animals pixels and Background pixels? This is a supervised learning task, however I cannot really map this task onto classification, or detection, or segmentation. This resembles clustering, however it is not, since it is supervised.
Is there a common approach for solving such problems?