Hello, we are using Nvidia pre-built containers, they work just fine with minor issues here and there that we were able to resolve, but we are wondering if there are any plans to update python in those containers from 3.6.9 (which is kinda legacy at this point). We ran into issues when we tried to do it ourselves and then build wheels for pytorch and opencv cuda enabled libraries and enable tensorRT support in those containers. As for the project, we are running yolov5 on our Jetson Xavier NX machine. Even tho it says it requires python >= 3.7, 3.6.9 version seems to do the job (we are using nvidia L4T r3.7.1 ML container for jetpack 4.6.1. These containers even work well for r3.7.2 (jetpack 4.6.2). At this point jetpack 5.0.1 is still in developer-preview version so we have stick to the 4.6.1/4.6.2.
Hi @DotaPie, we build the containers (and the wheels for PyTorch/TensorFlow/ect) for the default version of Python that comes with Ubuntu. On JetPack 4.x, that’s Ubuntu 18.04 and Python 3.6. On JetPack 5.x, that’s Ubuntu 20.04 and Python 3.8.
JetPack 5.0.2 is nearing release, so it will be out of developer preview soon.
Thank you for your swift answer! Looking forward for 5.0.2.