Conceptual

Feature Hierarchies from Edges to Object Parts in Deep Vision Networks

the hierarchy is what an encoder compresses into and what FID and CLIP both measure distances in