Read the original document by opening this link in a new tab.
Table of Contents
1. Introduction
2. Related Work
3. Approach
4. Building Efficient CNNs
5. Experiments
Summary
The document discusses the GhostNet, a novel approach to generate more feature maps from cheap operations in convolutional neural networks. It introduces the Ghost module, which utilizes linear transformations to create additional feature maps efficiently. By stacking Ghost modules in Ghost bottlenecks, the lightweight GhostNet architecture is established. Experimental results demonstrate the effectiveness of Ghost module in reducing computational costs while maintaining recognition performance. GhostNet outperforms state-of-the-art models like MobileNetV3 in various tasks with fast inference on mobile devices.