I installed pytorch1.11.0 and torchvision0.12.0 on AGX Orin, but it can't running successfully.
some error such as "functional.py:568: UserWarning: torch.meshgrid: in an upcoming release, it will be required to pass the indexing argument." "otImplementedError: Could not run 'torchvision::nms' with arguments from the 'CUDA' backend. This could be because the operator doesn't exist for this backend, or was omitted during the selective/custom build process (if using custom build). "
Is there anyone can fix it? thanks.
Hi,
Could you give our prebuilt package a try?
You can find the details in the below document:
For Orin, there are package for v.1.12.0 + JetPack 5.0 and v1.13.0+JetPack 5.0.2 (recommended).
Thanks.
I installed pytorchv1.13.0a0+08820cb0.nv22.7+torchvision0.13.0a0+da3794e+JetPack 5.0.2, and running the predict.py in the project(deep-learning-for-image-processing/pytorch_object_detection/mask_rcnn at master · WZMIAOMIAO/deep-learning-for-image-processing · GitHub),but there still some error(as Fig2). Do you know why? By the way , I’m the first time use Jetson applictions, so the more detailed it is ,the better.
And the other question is that, the program I generated on another computer(ubuntu 20 OS) with pyinstaller can’t running on the Orin(the error is : 1: Syntax error: “(” unexpected ). Do you know how to fix it?
Thanks a lot.
Hi,
The error is caused by some version compatibility issue.
Have you tried the sample in another environment before?
If yes, could you please share which PyTorch version is used with us?
About pyinstaller, not sure if the binary built with PyInstaller can be used across different platforms.
Especially, Jetson is an ARM system that is quite different from an x86 machine.
Is it possible to run the program with a python script on Orin directly?
Thanks.
I ran the sample on the win10(CPU is I5-11400 GPU is 2060Ti) and ubuntu22(CPU is I5-11400 GPU is 2060Ti) successfully. But I can’t run it on the Orin neither with the python script nor program produced by the pyinstaller. Than means I can’t running the program in any ways.This is the worst thing. By the way, I can’t use the Hello World AI either, there are some error on Cmake when I follow the guide provided by Nvidia.
So ,I think I need more detailed instructions. Or you can teach me how to restore the factory settings ,and then install the relevant packages step by step.
Thanks a lot.
Hi,
Is it possible to run the sample on Orin without wrapping with pyInstaller on Orin?
Or could you share some steps so we can test this in our environment?
Jetson uses ARM CPUs which are different from the desktop so the behavior will be different.
Thanks.
It can’t run the sample on Orin without wrapping with pyInstaller on Orin.
Now, I wonder if torch1.13.0a0+08820cb0.nv22.7 +torchvision 0.13.0a0+da3794e + JetPack 5.0.2 is right or wrong? Or you can tell me the right version and url for torch + torchvison +Jetpack. By the way, I just want to run the MaskRcnn from pytorch on Orin now. But there are always kinds of errors, and show that the version of torch and torchvision are not match.
Hi,
Please try the version shared in the below comment:
Thanks.
Yes, I can run the program successfully now. This problem has been bothering me for more than ten days .Thanks you very much. But the another question is that why the inference time on Orin is longer than PC( ubuntu22(CPU is I5-11400 GPU is 2060S)). I don’t think it should be.
Hi,
Have you maximized the device’s performance?
$ sudo nvpmodel -m 0
$ sudo jetson_clocks
Thanks.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.