A conditioned generator with cosine-similarity and feature-orthogonality losses inverts trained classifiers into diverse per-class images.
Reconstructing training data from multiclass neural networks, 2023
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Network Inversion and Its Applications
A conditioned generator with cosine-similarity and feature-orthogonality losses inverts trained classifiers into diverse per-class images.