VarGNet: Variable Group Convolutional Neural Network for Efficient Embedded Computing

TL;DR

VarGNet introduces a novel variable group convolution approach that simplifies hardware optimization and enhances efficiency across multiple vision tasks in embedded computing environments.

Contribution

The paper presents VarGNet, a new neural network design fixing channels in group convolutions, improving hardware optimization and efficiency over traditional fixed group number methods.

Findings

Demonstrated improved efficiency on embedded hardware

Achieved competitive accuracy across vision tasks

Simplified hardware implementation process

Abstract

In this paper, we propose a novel network design mechanism for efficient embedded computing. Inspired by the limited computing patterns, we propose to fix the number of channels in a group convolution, instead of the existing practice that fixing the total group numbers. Our solution based network, named Variable Group Convolutional Network (VarGNet), can be optimized easier on hardware side, due to the more unified computing schemes among the layers. Extensive experiments on various vision tasks, including classification, detection, pixel-wise parsing and face recognition, have demonstrated the practical value of our VarGNet.