TensorRT 3.0 RC now available with support for TensorFlow

A release candidate package for TensorRT 3.0 and cuDNN 7.0 is now available for Jetson TX1/TX2. Intended to be installed on top of an existing JetPack 3.1 installation, the TensorRT 3.0 RC provides the latest performance improvements and features:

  • TensorFlow 1.3 UFF (Universal Framework Format) importer
  • Upgrade from cuDNN 6 to cuDNN 7
  • New layers and parameter types

See the full Release Notes here and download the RC today from developer.nvidia.com/tensorrt

TensorRT development Python API - where can I find any information about this feature - documentation/code samples/tutorials/HowTo…?

Please see section (Caffe Python Workflow) TensorRT 3 User Guide (included in download) for docs and code examples of using the Python API.
Note that the Python API is currently supported on x86-only due to some .whl dependencies, see this item from the updated Release Notes:

TensorFlow UFF models can still be imported on ARM platforms from C++ using the NvUffParser.h API.

Thank you for your reply.

Does the Python API for TensorRT 3.0 RC gives us the flexibility to modify tensors? More specifically, are there functions which allow us to permute/reshape tensors similar to the following functions on pyTorch?
I wasn’t able to find them in the current documentation.

Hi, you should be able to use the Shuffle layer in network to permute/reshape tensors and accomplish this.

Thanks for your reply.
Unfortunately, there is no documentation about using the shuffle layer in python/doc/python3.5. It just says that the layer can be used to reshape or transpose data.
Is there any example/documentation which I can refer to?

How can I convert TensorFlow model into uff in python? The first step mentioned in the TensorRT3 User Guide is apt-get install python-tensorrt.

This step fails as it does not find the package.

Are you running that on PC (x86_64) or on Jetson?

The TensorRT Python API isn’t currently supported on Jetson/aarch64 because of a prerequisite’s dependency on Anaconda.

To convert the TensorFlow model into UFF from Python, please run these steps on PC. Then import the UFF on Jetson using TensorRT’s C++ API.

The JetPack 3.2 Developer Preview has been released for Jetson TX2 with support for TensorRT 3.0 RC2.

Please use JetPack 3.2 for automated installation of TensorRT 3.0 RC2, as opposed to the method here.


Can we install TensorRT-3.2RC on TX1?

If so, how? The link to TRT v3.0 will only download amd64 packages and not arm64.

The Jetpack needs to be installed on the host (amd64). After that connect the Jetson board to the host and the Jetpack will install all the required software on the Jetson.