DRIVE PX2 Tensorflow installation with TensorRT support

I have been trying to install TensorRT python bindings along with the Tensorflow r1.15.

This is how I am proceeding with (on host PC with existing installations from SDK for development)

  1. Firstly I get checkout my tensorflow to older versions because the Drive 5.0.10.3 comes with Cuda9.2, CUDNN 7.1.3 and TensorRT 4.0.0 on

  2. I am proceeding ahead with the tensorflow installation using cudnn support however without TensorRT support.

  3. Then after I install this, I tend to install the wheel which is already located at /usr/local/cuda-9.2/dl/tensorrt35.whl file

  4. Kindly throw some light if this is the right way.

I am finding it really hard to proceed any further as the support seems to be very limited with legacy version.

Also a quick question, has the support to the PX2 been stopped already ?
Does it support TensorRT5+, Cuda 10* and CUDNN latest versions for that matter ?

Thank you.

Dear @MaC345,
did you check the TF installation steps on DRIVE PX2 FAQ at Tensorflow installation on DPX2

Also a quick question, has the support to the PX2 been stopped already ?
Does it support TensorRT5+, Cuda 10* and CUDNN latest versions for that matter ?

Yes. There is more DRIVE SW releases targeted for DRIVE PX2.

Thank you for the info, I was able to make it work using my method generating wheel from existing sources. It worked out just fine !

Does the Drive PX2 support Tensorflow 2 as of yet ?

Dear @MaC345,

Does the Drive PX2 support Tensorflow 2 as of yet ?

No.

Glad to hear that you could install with your method. Could you share your installation steps on the thread to help community.

My method was found to be unstable and threw run time errors very, eventually I switched to yours. I guess I missed few corrections in the bazel.build.

Another quick question, does driveworks support UFF files, because most of your examples use caffee support. We have a code in tensorflow and we would like to port it to using UFF file to access lidar based data. Just an important note: the parsed UFF file will not have any external layer added over to it later using TensorRT inside the program.

Dear @MaC345,
Yes. DW supports UFF files too. You need to use TensorRT_Optimization tool to UFF → TRT engine conversion.