Tensor Reordering for CNN Compression

TL;DR

This paper introduces a spectral domain pruning method for CNNs using DCT, enabling significant compression with minimal accuracy loss by tensor reordering and fine-tuning.

Contribution

The novel approach applies DCT-based spectral pruning combined with tensor reordering to effectively compress CNNs while maintaining performance.

Findings

Achieves high layer compression with minor accuracy loss

Effective on ResNet-50 and MobileNet-V2 architectures

Minor fine-tuning recovers original performance

Abstract

We show how parameter redundancy in Convolutional Neural Network (CNN) filters can be effectively reduced by pruning in spectral domain. Specifically, the representation extracted via Discrete Cosine Transform (DCT) is more conducive for pruning than the original space. By relying on a combination of weight tensor reshaping and reordering we achieve high levels of layer compression with just minor accuracy loss. Our approach is applied to compress pretrained CNNs and we show that minor additional fine-tuning allows our method to recover the original model performance after a significant parameter reduction. We validate our approach on ResNet-50 and MobileNet-V2 architectures for ImageNet classification task.