Fizza Rubab

Face embeddings from different DNNs, even those trained on different data and with different architectures, can be surprisingly well-aligned with simple linear transformations, unlocking cross-model face recognition.