TensorRT-RepVGG
TensorRT-RepVGG copied to clipboard
upczww
TensorRT implementation of "RepVGG: Making VGG-style ConvNets Great Again"
Env GPU, Nano OS, Cuda version 10.2 TensorRT version About this repo repvgg Your problem sudo ./repvgg -s RepVGG-B2 // serialize model to plan file i.e. 'RepVGG-B2.engine' finished load weights...
Hi, what is the speed of your models: 1. before converting to VGG? 2. after converting to VGG ? 3. after converting to TensorRT ? Thx~
Hi, Thanks for the code. Why you need to write your own parser?, Why not converting to onnx and use TensorRT parser/network builder?. Because after converting the model to deploy...
repvgg.cpp 146 else layer = RepVGGBlock(network, weightMap, *layer->getOutput(0), inch, outch, 1, group, lname + std::to_string(i) + "."); here inch and outch are constant ,but when i>0 inch should be outch,...
