Architecture

Control engineering

Pattern recognition (psychology)

Decoupling (probability)

Inference

Residual neural network

FLOPS

Convolution (computer science)

Algorithm

Epistemology

Computer engineering

Engineering

Visual arts

Philosophy

Parallel computing

Artificial intelligence

Set (abstract data type)

Programming language

Code (set theory)

Computer science

Simple (philosophy)

Artificial neural network

Convolutional neural network

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 3 × 3 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.

RepVGG: Making VGG-style ConvNets Great Again