Xavier NX Jetpack 4.4 GA TRT gives wrong results for a specific caffe model

This is a follow up to

Xavier NX Jetpack 4.4 GA TRT gives wrong results for the caffe model res10_300x300_ssd_iter_140000_fp16.caffemodel.

This model runs correctly in Nano Jetpack 4.3

We have a self contained program to reproduce it here:

Instructions on how to run the program are in “readme.txt”

Hi,

Thanks for your reporting.
We are checking this issue and update more information with you later.

Thanks.

Hi,

Confirmed that this issue can be reproduced in our environment.
We have passed this to our internal team.

Will update more information once we got any progress.
Thanks.

Hi,

Thanks for your patience.

This issue can be fixed by replacing the TensorRT plugin library from the OSS GitHub.
Here are the detail steps:

1. Update cmake

$ wget https://github.com/Kitware/CMake/releases/download/v3.13.5/cmake-3.13.5.tar.gz
$ tar xvf cmake-3.13.5.tar.gz
$ cd cmake-3.13.5/
$ ./configure
$ make -j 8
$ sudo make install
$ cd ..

2. Build OSS

$ git clone -b release/7.1 https://github.com/nvidia/TensorRT
$ cd TensorRT
$ git submodule update --init --recursive
$ mkdir -p build && cd build
$ /usr/local/bin/cmake .. -DGPU_ARCHS=72  -DTRT_LIB_DIR=/usr/lib/aarch64-linux-gnu/ -DCMAKE_C_COMPILER=/usr/bin/gcc -DTRT_BIN_DIR=`pwd`/out
$ make nvinfer_plugin -j$(nproc)

3. Replace libnvinfer_plugin.so.7.1.3 library

$ sudo mv /usr/lib/aarch64-linux-gnu/libnvinfer_plugin.so.7.1.3 ${HOME}/libnvinfer_plugin.so.7.1.3.bak
$ sudo cp libnvinfer_plugin.so.7.1.3 /usr/lib/aarch64-linux-gnu/libnvinfer_plugin.so.7.1.3

Thanks.

1 Like

Many thanks!