ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices

By X. Zhang et al

Read the original document by opening this link in a new tab.

Table of Contents

1. Introduction
2. Related Work
3. Approach
4. Experiments
4.1. Ablation Study
4.1.1 Pointwise Group Convolutions
4.1.2 Channel Shuffle vs. No Shuffle
4.2. Comparison with Other Structure Units

Summary

ShuffleNet is introduced as an extremely computation-efficient CNN architecture designed for mobile devices with limited computing power. The architecture utilizes pointwise group convolution and channel shuffle operations to reduce computation cost while maintaining accuracy. Experimental results demonstrate superior performance over existing structures, especially in the context of small networks. The paper discusses the importance of pointwise group convolutions and the channel shuffle operation in enhancing classification performance. Overall, ShuffleNet proves to be a promising solution for efficient neural networks on mobile platforms.