Built independently by an author, for readers. Read the story and support ChapterPal
ChapterPal Logo
Log inSign up

keyword

ShuffleNet unit

A ShuffleNet unit is a computationally efficient building block for convolutional neural networks designed to achieve high accuracy and fast processing speeds on resource-constrained mobile and embedded devices. Derived from bottleneck and residual network designs, the unit minimizes computational complexity and parameter count by replacing standard dense convolutions with pointwise group convolutions and depthwise convolutions. To overcome the restriction where group convolutions isolate channel subsets and impede feature communication, the unit introduces a channel shuffle operation that systematically permutes and mixes feature channels across different groups. Across different architectural variants, the unit may also incorporate channel splitting, concatenation, and downsampling shortcuts to optimize memory access costs and maintain rich feature representations during real-time inference.

1 item

ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design

ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design

Ningning Ma, Xiangyu Zhang, Haitao Zheng, Jian Sun

OrganizationsMegvii TechnologyTsinghua University

Why you should read this

Establishes practical convolutional neural network design principles by evaluating real-world inference speed and memory access costs rather than theoretical FLOPs, introducing the highly efficient ShuffleNet V2 architecture.

Currently, the neural network architecture design is mostly guided by the \emph{indirect} metric of computation complexity, i.e., FLOPs. However, the \emph{direct} metric, e.g., speed, also depends on the other factors such as memory access cost and platform characterics. Thus, this work proposes to evaluate the direct metric on the target platform, beyond only considering FLOPs. Based on a series of controlled experiments, this work derives several practical \emph{guidelines} for efficient network design. Accordingly, a new architecture is presented, called \emph{ShuffleNet V2}. Comprehensive ablation experiments verify that our model is the state-of-the-art in terms of speed and accuracy tradeoff.

Added

2026-09-09