Network Decoupling: From Regular to Depthwise Separable Convolutions

TL;DR

This paper introduces network decoupling, a training-free method that approximates regular convolutions with depthwise separable convolutions, enabling significant CNN speedups with minimal accuracy loss.

Contribution

It mathematically relates regular and depthwise separable convolutions and proposes a training-free decoupling method for faster CNN inference.

Findings

ND achieves about 2X speedup on VGG16

Combining ND with other methods yields 3.7X speedup

ND is applicable to various network architectures

Abstract

Depthwise separable convolution has shown great efficiency in network design, but requires time-consuming training procedure with full training-set available. This paper first analyzes the mathematical relationship between regular convolutions and depthwise separable convolutions, and proves that the former one could be approximated with the latter one in closed form. We show depthwise separable convolutions are principal components of regular convolutions. And then we propose network decoupling (ND), a training-free method to accelerate convolutional neural networks (CNNs) by transferring pre-trained CNN models into the MobileNet-like depthwise separable convolution structure, with a promising speedup yet negligible accuracy loss. We further verify through experiments that the proposed method is orthogonal to other training-free methods like channel decomposition, spatial decomposition, etc. Combining the proposed method with them will bring even larger CNN speedup. For instance, ND itself achieves about 2X speedup for the widely used VGG16, and combined with other methods, it reaches 3.7X speedup with graceful accuracy degradation. We demonstrate that ND is widely applicable to classification networks like ResNet, and object detection network like SSD300.