Harmonic Convolutional Networks based on Discrete Cosine Transform

18 Jan 2020Matej UlicnyVladimir A. KrylovRozenn Dahyot

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)

PDF Abstract

Evaluation Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK COMPARE
Image Classification ImageNet Harm-SE-RNX-101 64x4d (320x320, Mean-Max Pooling) Top 1 Accuracy 82.66% # 34
Image Classification ImageNet Harm-SE-RNX-101 64x4d (320x320, Mean-Max Pooling) Top 5 Accuracy 96.29% # 24
Image Classification ImageNet Harm-SE-RNX-101 64x4d (320x320, Mean-Max Pooling) Number of params 88.2M # 1