weixin_45415546的博客

升级到Tensort7, ONNX模型创建网络和预测都需要用V2版本。创建模型 const auto explicitBatch = 1U << static_cast<uint32_t>(NetworkDefinitionCreationFlag::kEXPLICIT_BATCH); INetworkDefinition *network = builder->createNetworkV2(explicitBatch); // 创建TensorRT网络

1.Convolutionout = floor((in+2padding-dilation(kernel_size-1)-1)/stride+1)dilation默认情况为１out=floor((in+2padding-kernel_size)/stride+1)2.Poolingout = floor((in+2padding-dilation(kernel_size-1)-1)/s...

Concat是将多个bottom合并成一个topslice层把一个bottom分解成多个topslice层三个参数:axis:默认为1，切分channel维度slice_dim:过时参数同axis,不支持负数slice_point：切分点,没有设置的时候则对axis进行均匀分解message SliceParameter { // The axis along which to s...