Harmonic Convolutional Networks based on Discrete Cosine Transform

Matej UlicnyVladimir A. KrylovRozenn Dahyot

   Papers with code   Abstract  PDF

Convolutional neural networks (CNNs) learn filters in order to capture local correlation patterns in feature space. In this paper we propose to revert to learning combinations of preset spectral filters by switching to CNNs with harmonic blocks... (read more)

Benchmarked Models

RANK
MODEL
REPO
CODE RESULT
PAPER RESULT
ε-REPRODUCED
BUILD
1
Harm-SE-RNX-101 64x4d
(320x320, Mean-Max Poo…)
82.9%
82.7%
2
Harm-SE-RNX(64x4d)
(320x320, Mean-Max Poo…)
82.9%
--
3
Harm-SE-RNX-101 64x4d
81.6%
81.6%
5
Harm-SE-RNX-101 32x4d
(320x320, Mean-Max Poo…)
81.5%
81.3%
6
Harm-SE-RNX(32x4d)
(320x320, Mean-Max Poo…)
81.5%
--
7
Harm-SE-RNX-101 32x4d
80.6%
80.5%