RepVGG: Making VGG-style ConvNets Great Again

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... (read more)

PDF Abstract
TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Semantic Segmentation Cityscapes val RepVGG-B2 mIoU 80.57% # 5
Image Classification ImageNet RepVGG-B2g4 Top 1 Accuracy 78.5% # 100
Number of params 55.77M # 32
Image Classification ImageNet RepVGG-B2 Top 1 Accuracy 78.78% # 95
Number of params 80.31M # 21

Methods used in the Paper