Table Of Contents

• Description
• How does this sample work?
• Prerequisites
• Running the sample
  • Cloning the packnet repository
  • Conversion to ONNX
  • Inference with TensorRT
  • Sample --help options
• Additional resources
• License
• Changelog
• Known issues

This sample, samplePackNet, is a Python sample which uses TensorRT to perform inference with PackNet network. PackNet is a self-supervised monocular depth estimation network used in autonomous driving.

This sample converts the Pytorch graph into ONNX and uses ONNX-parser included in TensorRT to parse the ONNX graph. The sample also demonstrates

• Use of custom layers (plugins) in ONNX graph. These plugins would be automatically registered in TensorRT by using REGISTER_TENSORRT_PLUGIN API.
• Use of ONNX-graphsurgeon (ONNX-GS) API to modify layers or subgraphs in the ONNX graph. For this network, we transform Group Normalization, upsample and pad layers to remove unnecessary nodes for inference with TensorRT.

1. Upgrade pip version and install the sample dependencies.

   pip3 install --upgrade pip
   pip3 install -r requirements.txt

On PowerPC systems, you will need to manually install PyTorch using IBM's PowerAI.

Clone the packnet repository and update PYTHONPATH.

git clone https://github.com/TRI-ML/packnet-sfm.git packnet-sfm
pushd packnet-sfm && git checkout tags/v0.1.2 && popd
export PYTHONPATH=$PWD/packnet-sfm # Note on Windows, the export command is: set PYTHONPATH=%cd%\packnet-sfm

Run the following command to convert the Packnet pytorch network to ONNX graph. This step also includes handling custom layers (Group Normalization) and using ONNX-GS to modify upsample and pad layers.

python3 convert_to_onnx.py --output model.onnx

Once the ONNX graph is generated, use trtexec tool (located in bin directory of TensorRT package) to perform inference on a random input image.

trtexec --onnx=model.onnx

Please refer to trtexec tool for more commandline options.

To see the full list of available options and their descriptions, use the -h or --help command line option. For example:

convert_to_onnx.py -h

The following resources provide a deeper understanding about PackNet network and importing a model into TensorRT using Python:

PackNet

• 3D Packing for Self-Supervised Monocular Depth Estimation
• TRI-ML Monocular Depth Estimation Repository

Parsers

• ONNX Parser

Documentation

• Introduction To NVIDIA’s TensorRT Samples
• Working With TensorRT Using The Python API
• Importing A Model Using A Parser In Python
• NVIDIA’s TensorRT Documentation Library

For terms and conditions for use, reproduction, and distribution, see the TensorRT Software License Agreement documentation.

August 2025:

• Removed support for Python versions < 3.10.

August 2023:

• Update ONNX version support to 1.14.0
• Removed support for Python versions < 3.8. August 2021: Update sample to work with latest torch version June 2020: Initial release of this sample

There are no known issues in this sample