20210208-mmdetection模型转为tensorrt(trt)

1. 下载下载对应的的cuda，由于本机装的是10.1，所以下了个cuda_10.0.130_410.48_linux.run

2. 下载配置tensorrt，我下的是：TensorRT-7.0.0.11，配置如下：

export TR_PATH=/path_to_trt/TensorRT-7.0.0.11
export PATH=$PATH:$TR_PATH/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$TR_PATH/lib
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$TR_PATH/targets/x86_64-linux-gnu/lib

3. 安装pucuda

4. 安装tensorrt-7.0.0.11-cp37-none-linux_x86_64.whl，因为我的python版本是3.7，文件在TensorRT-7.0.0.11/python下

3. 安装mmdetection-to-tensorrt：地址：GitHub - grimoire/mmdetection-to-tensorrt: convert mmdetection model to tensorrt, support fp16, int8, batch input, dynamic shape etc.，按照说明安装，可能会缺少库，如下

3.1 torch2trt dynamic : GitHub - grimoire/torch2trt_dynamic: A pytorch to tensorrt convert with dynamic shape support

3.2 amirstan_plugin : https://github.com/grimoire/amirstan_plugin