I am trying to provide support for YOLO model in Yocto environment. There is already a good amount of Yocto support available in GitHub - OE4T/meta-tegra: BSP layer for NVIDIA Jetson platforms, based on L4T repository, however Python library versions (3.6) provided by NVIDIA do not match our distro Python version (3.9)
What I basically need is to integrate “tensorrt” Python library to Yocto. That requires installing python3-libnvinfer_7.1.3-1+cuda10.2_arm64.deb. which currently has Python3.6 bindings and creates /usr/lib/python3.6/dist-packages/tensorrt.
If there is a way if I could get Python3.9 binded version of this package, it would greately boost our work.
Sorry that we only have the python 3.6 package since it is the default python in Linux 18.04.
But since TensorRT also has C++ library, you may create the binding on your own.
Hi, I was able to compile .whl file thanks to this repository: GitHub - NVIDIA/TensorRT: TensorRT is a C++ library for high performance inference on NVIDIA GPUs and deep learning accelerators.
However, only 7.2 version has python/ folder. I need to do it for 184.108.40.206 version. Since there is no python/ folder in 7.1 branch, how do I compile wheel file for that?
Thanks for the support.
Sorry that the python folder is added in v7.2 and it doesn’t be backported to v7.1.3.
Maybe you can check if it is possible to use the CMakefile of v7.2 with some modification.
Nevermind, this still works for me. Thanks for the help.