CAFFE to TensorRT Model Tool

• Overview
• Building and Running
• Key Structure and Classes

---

Overview

NVIDIA^® TensorRT^™ is an accelerated neural network inference engine and run-time library. ConvertCaffeToTrtModel is a standalone model conversion tool that converts a CAFFE network to a TensorRT compatible model. This tool runs offline on the NVIDIA^® Jetson^™ platform and provides a cached TensorRT model stream to prevent subsequent repetitive network conversion. Using this converted model, TensorRT-based applications can improve greatly in accuracy and performance.

If the source model changes (i.e., is retrained), the tool performs conversion again to enable TensorRT accelerated inference.

Building and Running

Prerequisites

• You have followed Steps 1-3 in Building and Running.
• If you are building from your host Linux PC (x86), you have followed Step 4 in Building and Running.
• You have installed the TensorRT package.
• You have a trained a deep-learning network.

To build:

TBD

To run

• Enter:

  ./ConvertCaffeToTrtModel -n ../../data/Model/GoogleNet_one_class/GoogleNet_modified_oneClass_halfHD.prototxt \ -l ../../data/Model/GoogleNet_one_class/GoogleNet_modified_oneClass_halfHD.caffemodel \ -m detection -o coverage,bboxes -f fp16 -b 2 -w 115343360 -s trtModel.cache

To get a list of supported options

• Use the -h option.

---

Key Structure and Classes

The CudaEngine structure is a TensorRT interface that invokes the TensorRT function.

The sample uses the following function:

Function	Description
caffeToTRTModel	Uses TensorRT API to transfer a network model from CAFFE to TensorRT.