tensorRT混合精度

精度类型：

DataType::kFLOAT, DataType::kHALF and DataType::kINT32  DataType::kINT8

指定特定层的精度：

layer->setPrecision(nvinfer1::DataType::kINT8)

如何查找特定层：

void PrintLayerInfo(nvinfer1::INetworkDefinition *network)

{

std::cout << "------Network layers------" << std::endl;

int layer_num = network->getNbLayers();

nvinfer1::ILayer *layer;

int layer_index;

for(int i=0; i<layer_num; i++)

{

layer = network->getLayer(i);

layer_index = static_cast<int>(layer->getType());

switch(layer_index)

{

case 0:

{

......

}

}

}

}

参考例子：

https://codeleading.com/article/6074265133/

各网络层对应的编号：
搜索LayerType
https://docs.nvidia.com/deeplearning/tensorrt/api/c_api/_nv_infer_8h.html