Post Go back to editing

Deploy custom deep learning models on AD-3DSMARTCAM1-PRZ


I have difficulties deploying my own deep learning models on the AD-3DSMARTCAM1-PRZ.

For one thing, the free disk space of the camera is so small, which is only around 1.3G available without installing any thing other than aditof SDK.

It is impossible to install Pytorch, cuDNN or whatever needed for my own deep learning applications.

For another, it seems that the jetson nano board on the camera is already installed JetPack (it shows the JetPack version is 4.5).

So I assume if the JetPack is already installed, the development environment would have been provided.

But I found that the installation is kind of incomplete. e.q. there is no TensorRT, no "lib" directory under "/usr/local/cuda-10.2"...

Therefore, my question is what is the recommended way to deploy our own deep learning applications on  AD-3DSMARTCAM1-PRZ?

Do I need docker container or something?

I would be very appreciate if you can help me with this.



Edit subject
[edited by: r09921135 at 2:06 PM (GMT -4) on 25 Aug 2022]
Parents Reply Children
  • In case of an existing deep learning model, it can be implemented either in tensor flow or in other frameworks that support CUDA API. This can be implemented on the camera in order to do the processing. There is already an example on a branch of the repository for this implementation. But for development (especially with graphical aids) a more performant device is advised to be used.