Add IQR Proof Of Concept
We want to add support for IQR, either using team features, or fusion features (likely the latter). The modularized repo for SMQTK IQR is: https://github.com/Kitware/SMQTK-IQR but it needs some TLC, which we should provide.
On our side, we need to add one additional feature, which is the ability to output intermediate network activations as features from a fusion network. This should not be a big lift. We need to modify the multimodal transformer to provide these in its output dictionary (likely based on some request at predict time) and then create a CocoStitcher to combine them into a multi-channel image.
From there we need SQMTK to ingest these features as descriptors and then build out the proof of concept.