Scale-equivariant (first-order homogeneous) network components are proposed as an inductive bias that lets denoisers trained on uniform Gaussian noise generalize to spatially varying OOD noise.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Towards Robust Image Denoising with Scale Equivariance
Scale-equivariant (first-order homogeneous) network components are proposed as an inductive bias that lets denoisers trained on uniform Gaussian noise generalize to spatially varying OOD noise.