RepVGG

论文：https://arxiv.org/abs/2101.0369
代码：

文章结构概览

• 1. Motivation

  • 2. Contribution

  • 3. Building RepVGG via Structural Re-param

  • • 3.1 Simple is Fast, Memory-economical, Flexible
    • 3.2 Traing-time Multi-branch Architecture
    • 3.3 Architectural Specification

  • 4.Experiments

  • • 4.1 RepVGG for ImageNet Classification
    • 4.2 Structural Re-parameterization is the Key
    • 4.3 Semantic Segmentation

  • Reference

1. Motivation

当前更为复杂的卷积网络能够实