TensorRT backend for ONNX

Parses ONNX models for execution with TensorRT.

See also the TensorRT documentation.

ONNX Python backend usage

The TensorRT backend for ONNX can be used in Python as follows:

import onnx
import onnx_tensorrt.backend as backend
import numpy as np

model = onnx.load("/path/to/model.onnx")
engine = backend.prepare(model, device='CUDA:1')
input_data = np.random.random(size=(32, 3, 224, 224)).astype(np.float32)
output_data = engine.run(input_data)[0]
print(output_data)
print(output_data.shape)

Executable usage

ONNX models can be converted to serialized TensorRT engines using the onnx2trt executable:

onnx2trt my_model.onnx -o my_engine.trt

ONNX models can also be converted to human-readable text:

onnx2trt my_model.onnx -t my_model.onnx.txt

See more usage information by running:

onnx2trt -h

C++ library usage

The model parser library, libnvonnxparser.so, has a C++ API declared in this header:

NvOnnxParser.h

TensorRT engines built using this parser must use the plugin factory provided in libnvonnxparser_runtime.so, which has a C++ API declared in this header:

NvOnnxParserRuntime.h

Installation

Dependencies

• Protobuf
• TensorRT 3+

Download the code

Clone the code from GitHub.

git clone --recursive https://github.com/onnx/onnx-tensorrt.git

Executable and libraries

Suppose your TensorRT library is located at /opt/tensorrt. Build the onnx2trt executable and the libnvonnxparser* libraries using CMake. Note that onnx-tensorrt will by default tell the CUDA compiler generate code for the latest SM version. If you are using a GPU with a lower SM version you can specify which SMs to build for by using the optional -DGPU_ARCHS flag. For example, if you are running TensorRT on an older Pascal card such as a GTX 1080, you can specify -DGPU_ARCHS="61" to generate CUDA code specifically for your card.

See here for finding what maximum compute capability your specific GPU supports.

mkdir build
cd build
cmake .. -DTENSORRT_ROOT=/opt/tensorrt
OR
cmake .. -DTENSORRT_ROOT=/opt/tensorrt -DGPU_ARCHS="61"
make -j8
sudo make install

Python modules

Python bindings for the ONNX-TensorRT parser in TensorRT versions >= 5.0 are packaged in the shipped .whl files. No extra install is necessary.

For earlier versions of TensorRT, the Python wrappers are built using SWIG. Build the Python wrappers and modules by running:

python setup.py build
sudo python setup.py install

Docker image

Build the onnx_tensorrt Docker image by running:

cp /path/to/TensorRT-3.0.*.tar.gz .
docker build -t onnx_tensorrt .

Tests

After installation (or inside the Docker container), ONNX backend tests can be run as follows:

Real model tests only:

python onnx_backend_test.py OnnxBackendRealModelTest

All tests:

python onnx_backend_test.py

You can use -v flag to make output more verbose.

Pre-trained models

Pre-trained models in ONNX format can be found at the ONNX Model Zoo