Efficient CNN Architecture Design Guided by Visualization

TL;DR

This paper introduces VGNetG, a new CNN architecture guided by visualization insights, achieving better accuracy and efficiency than previous models, and explores replacing learnable convolutions with fixed edge detectors.

Contribution

The paper proposes visualization-guided design guidelines for CNNs, leading to a parameter-efficient architecture with improved accuracy and speed, and demonstrates replacing learnable layers with fixed edge detection kernels.

Findings

VGNetG achieves 67.7% top-1 accuracy with 0.99M parameters.

Parameter reduction of 30-50% compared to previous networks.

Replacing learnable convolutions with edge detectors maintains competitive accuracy.

Abstract

Modern efficient Convolutional Neural Networks(CNNs) always use Depthwise Separable Convolutions(DSCs) and Neural Architecture Search(NAS) to reduce the number of parameters and the computational complexity. But some inherent characteristics of networks are overlooked. Inspired by visualizing feature maps and N$\times$N(N$>$1) convolution kernels, several guidelines are introduced in this paper to further improve parameter efficiency and inference speed. Based on these guidelines, our parameter-efficient CNN architecture, called \textit{VGNetG}, achieves better accuracy and lower latency than previous networks with about 30%$\thicksim$50% parameters reduction. Our VGNetG-1.0MP achieves 67.7% top-1 accuracy with 0.99M parameters and 69.2% top-1 accuracy with 1.14M parameters on ImageNet classification dataset. Furthermore, we demonstrate that edge detectors can replace learnable depthwise convolution layers to mix features by replacing the N$\times$N kernels with fixed edge detection kernels. And our VGNetF-1.5MP archives 64.4%(-3.2%) top-1 accuracy and 66.2%(-1.4%) top-1 accuracy with additional Gaussian kernels.