Jetson Nano L4T 32.3.1 and yolov7 compatibility

[guillermo.velc]

Hello, i have an issue using yolov7 in the Jetpack 4.3 L4T 32.3.1.
It seems that yolov7 requires torch.cuda.amp library, which was introduced in pytorch 1.5.0, but if i try to install any version grater than pytorch 1.4 i got an error:
OSError: libcurand.so.10: cannot open shared object file: No such file or directory as soon as i import torch
And as i have found, this is due to the Jetpack’s version. I know the easiest solution is to upgrade the Jetpack version but since im working remotely i would like to know if there is any other solution i can try before.

[AastaLLL]

Hi,

Is your PyTorch package built with r32.3.1?
If not, please try to install one.

You can find more information in the below topic:

PyTorch for Jetson

Below are pre-built PyTorch pip wheel installers for Jetson Nano, TX1/TX2, Xavier, and Orin with JetPack 4.2 and newer. Download one of the PyTorch binaries from below for your version of JetPack, and see the installation instructions to run on your Jetson. These pip wheels are built for ARM aarch64 architecture, so run these commands on your Jetson (not on a host PC). You can also use the pre-built l4t-pytorch and l4t-ml container images and Dockerfiles. PyTorch pip wheels JetPack 5

PyTo…

Thanks.

[guillermo.velc]

Hi, thank you for you reply

The latest version of PyTorch i was able to use with no problems was 1.4, which according to the topic you mentioned is compatible with my version of JetPack (4.3). But the problem is that to use yolov7 i need a greater version.

UPDATE:
I updated my JetPack version following the steps in this page to JetPack 4.4.1 [L4T 32.4.4], so i was able to install a newer version of PyTorch (1.8), but im still having errors.

As i have read, cuda 10.2 should solve this problem, but it could be because JetPack 4.4.1 hasn’t been installed natively.

I have made sure to modify the symbolic link to the latest version of cuda (i have tried both)

The situation is that i need to create a compiled program that uses yolov7 for a TX2 (remote) using my Nano (local), but when i compile using newer JetPack and PyTorch versions on the Nano, it doesn’t work in TX2. So i would like to find a solution that does not involve reflashing the card.

Thanks.

[AastaLLL]

Hi,

Could you try if torch-1.10.0-cp36-cp36m-linux_aarch64.whl works?
It is the latest release for JetPack 4.4.

Thanks.