RepVGG: Making VGG-style ConvNets Great Again

TL;DR

RepVGG introduces a re-parameterization technique enabling a VGG-like inference architecture with multi-branch training, achieving high accuracy and fast inference on ImageNet, outperforming ResNet and competing with modern models.

Contribution

The paper proposes RepVGG, a novel CNN architecture that decouples training and inference structures via structural re-parameterization, leading to improved accuracy and speed.

Findings

Achieves over 80% top-1 accuracy on ImageNet.

Runs 83% faster than ResNet-50 on NVIDIA 1080Ti.

Outperforms ResNet and rivals EfficientNet and RegNet in accuracy-speed trade-off.

Abstract

We present a simple but powerful architecture of convolutional neural network, which has a VGG-like inference-time body composed of nothing but a stack of 3x3 convolution and ReLU, while the training-time model has a multi-branch topology. Such decoupling of the training-time and inference-time architecture is realized by a structural re-parameterization technique so that the model is named RepVGG. On ImageNet, RepVGG reaches over 80% top-1 accuracy, which is the first time for a plain model, to the best of our knowledge. On NVIDIA 1080Ti GPU, RepVGG models run 83% faster than ResNet-50 or 101% faster than ResNet-101 with higher accuracy and show favorable accuracy-speed trade-off compared to the state-of-the-art models like EfficientNet and RegNet. The code and trained models are available at https://github.com/megvii-model/RepVGG.