Uniform INT8 quantization of learned image codecs becomes more accurate when tensors are first re-expressed in a layer-adaptive Hadamard basis.
End-to-end learned image compression with fixed point weight quantization,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
eess.IV 1years
2026 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Hadamard-Domain Model Quantization for Learned Image Coding
Uniform INT8 quantization of learned image codecs becomes more accurate when tensors are first re-expressed in a layer-adaptive Hadamard basis.