Problems with sampleMNIST on Jetson Nano

I am having problems running the TensorRT example sampleMNIST on the Jetson Nano.

First, I am unble to build samples as described in GitHub - NVIDIA/TensorRT at release/6.0, “Building the TensorRT Components”, this requires CMake version 3.13 or later, but the version provided with the SD image is only 3.10.2 which is the latest available via ‘apt install’.

I was able to create my own Makefile in order to build sampleMNIST, but this generates a segmentation fault when trying create the builder config on line 156 of sampleMNIST.cpp

auto config = SampleUniquePtr<nvinfer1::IBuilderConfig>(builder->createBuilderConfig());

Similarly, if I follow the instructions from https://docs.nvidia.com/deeplearning/sdk/tensorrt-developer-guide/index.html#import_caffe_c, I also get a seg fault when calling createBuilderConfig().

Just FYI, may not be the same as your case.

I also hit segfault at this exact API call. It turned out I was using a newer API than the system installed on, like using TensorRT 5.1 API on a system with TensorRT 5.0 installed, cannot remember the exact versions though. Mysteriously the compiler was happy but apparently it won’t work at runtime.

Check the /usr/src dir, see if there’s a tensorrt folder inside. Using the sample code under that folder if present instead of the one in github.

Hi,

Please try the samples that are part of TRT package.
/samples/

Refer to below link for step by step instruction to run the sample:
https://github.com/NVIDIA/TensorRT/blob/release/6.0/samples/opensource/sampleMNIST/README.md#running-the-sample

Thanks