Closed doodieb15 closed 5 months ago
Python and GPU support on Jetson is a mess! I managed to install ObjectdetectionYOLOv5.3.1 with GPU support and ALPR without GPU support. Both are running stable now, detection rates of the Objectdetection plugin are rather high.
Start with the Ubuntu 20.04 image from QEngineering.
I am using a SDD as swap drive, I am quite sure this is needed.
Install python3.8-venv
For YOLOv5.3.1 use Python 3.8 and torch==1.13 and torchvision==1.14 from the Qengineering Website. Install both in the venv. For ALPR use Python 3.8 and paddlepaddle==2.5.2 CPU version. Install in the venv
I'm, not 100% sure though whether this was all that is needed, and I am tired to start all over again.
Area of Concern
Describe the bug Trying to install CodeprojectAI on the Jetson Nano 4GB following this guide: https://www.rootadmin.com/AI/docs/install/install_on_jetson.html.
Expected behavior Expect to be able to build
Screenshots Multiple errors during 'setup.sh'. i.e.: ALPR: ImportError: libcudart.so.11.0: cannot open shared object file: No such file or directory ObjectDetectionCoral 2.1.0: Self test: ./setup.sh: line 298: 37123 Illegal instruction (core dumped) "$venvPythonCmdPath" "$moduleStartFilePath" --selftest > /dev/null Self-test failed ObjectDetectionYOLOv5-3.1 1.9.0: ModuleNotFoundError: No module named 'torchvision'
Build does fail also
Your System (please complete the following information):
Additional context Add any other context about the problem here.