Python [Anaconda] on TX1

I was looking into how to get started with GPU Accelerated Computing with Python on the TX1. It seems like Anaconda is the recommended option, but it does not come with the JetPack.

The NVidia page links to the anaconda download page, but its seems like there are no fitting packages for the 64-bit ARM CPU of the TX1.

Can someone link me to a guide and let me know how else I could interface with CUDA and openCV from Python?

Thank you!

For me, the python package is installed automatically after flashing Jetpack. Running ‘python’ from terminal provides the interactive python prompt. If you want to find all the packages available with python, try using the apt-cache tool (apt-cache search python). You can try installing libopencv4tegra package for the bindings

(Ah, I was finally able to log in after creating a separate account!)


I read this response of yours and simply had to respond.

I have the Jetson board and am working with Caffe.

It is true that the python package is installed in the pristine Jetson. It is adequate for building Caffe and running the basic demos.

HOWEVER, to do anything powerful like training, you will find yourself in need of progressively more python libraries and will eventually hit a brick wall - I forget which particular library it was - because of lack of support for the ARM.

The Caffe documentation recommends the Anaconda python build from the get-go, and I ended up breaking even my working examples trying to get Anaconda to run - and subsequently removing it when it didn’t hurt. (A 1.5 month hiatus due to a major life change didn’t help, but that’s besides the point.)

I would very much like to know how to install Anaconda on Jetson.

From what I understand, for most CV-type deep learning tasks, you don’t want to do training on a Jetson (unless you want to waste time/cycles), instead you’d want to train on a full desktop machine then transfer the model to the Jetson; see this thread.