Can Representation Learning for Multimodal Image Registration be Improved by Supervision of Intermediate Layers?

Multimodal imaging and correlative analysis typically require image alignment. Contrastive learning can generate representations of multimodal images, reducing the challenging task registration to a monomodal one. Previously, additional supervision on intermediate layers in contrastive has improved biomedical classification. We evaluate if similar approach improves learned for boost performance. explore three approaches add latent features bottleneck layer U-Nets encoding images different critic functions. Our results show that without perform best downstream two public datasets. investigate performance drop by exploiting recent insights classification self-supervised learning. visualize spatial relations means multidimensional scaling, lead partial dimensional collapse embedding space.