Harmonic Convolutional Networks based on Discrete Cosine Transform
This work addresses the need for more efficient and compressible CNN architectures for computer vision tasks, though it appears incremental as it builds on existing models by replacing layers with DCT-based blocks.
The authors tackled the problem of learning filters in convolutional neural networks by proposing harmonic blocks based on the Discrete Cosine Transform (DCT), which replace conventional layers and enable efficient compression through truncation of high-frequency information. They demonstrated benefits in image classification, object detection, and semantic segmentation applications with extensive experimental validation.
Convolutional neural networks (CNNs) learn filters in order to capture local correlation patterns in feature space. We propose to learn these filters as combinations of preset spectral filters defined by the Discrete Cosine Transform (DCT). Our proposed DCT-based harmonic blocks replace conventional convolutional layers to produce partially or fully harmonic versions of new or existing CNN architectures. Using DCT energy compaction properties, we demonstrate how the harmonic networks can be efficiently compressed by truncating high-frequency information in harmonic blocks thanks to the redundancies in the spectral domain. We report extensive experimental validation demonstrating benefits of the introduction of harmonic blocks into state-of-the-art CNN models in image classification, object detection and semantic segmentation applications.