Hi everyone,

I’m working on a project that utilizes multiple classifiers, each trained on a distinct subset of classes. These classifiers are intended to tackle various aspects of the classification process, but I’m encountering difficulties in combining their outputs into a single cohesive prediction.

For instance, if one classifier is tasked with differentiating between classes 0 and 1, while another focuses on classes 2 and 3, how can we effectively merge their results when the correct prediction is class 1? Initially, we attempted to use an "other" class to signify when an input doesn’t fit within a classifier’s specified classes, but this approach hasn’t produced satisfactory outcomes.

We are now considering adding an extra head for detecting out-of-distribution classes, but we seek a more efficient and streamlined method. Has anyone faced a similar challenge or have suggestions for effectively aggregating outputs from multiple classifiers?

I appreciate any insights you can share!