Can I install DIGITS on Jetson?

Hi,

    Can I install DIGITS or DIGITS2 on Jetson? Can you share how to install DIGITS? Thank you! 

One experiment I did was to start installing DIGITS2 on a TK1. It almost seemed to work…then it got to an x86 64-bit binary which it could not execute. This is “venv/bin/python”, so I removed this from the unpacked install tree, and used a symbolic link to the native /usr/bin/python. Install completed.

After this the “runtests.sh” failed, although this seemed to possibly be setup of related packages…i.e., it might work after adjusting other packages (or might not). I would expect the TX1 would be similar.

Did anyone get DIGITs to work on the Jetson TX1?
This post from a related thread (regarding the Jetson TK1) seems to indicate that Digits isn’t supported on ARM (at least at the time of that posting in 2015).
[url]https://devtalk.nvidia.com/default/topic/866425/jetson-tk1/digits-on-jetson-tk1-create-db-error-11-/post/4645163/#4645163[/url]

Hello Greendogo, the Jetson platform is not a supported platform for DIGITS at this time.

strange… i have it running on the jetson…

Which L4T version?

head -n 1 /etc/nv_tegra_release

Could you please also tell us if you did it with Tegra TK1 or TX1?

Hi, is it still the case? Can I run nvidia digits 5 on a jetson tx2 ?

Cheers,

Ugo

I am also trying to get DIGITS on my Jetson TX1 or a workaround.

DIGITS consists of a series of Python scripts, which you could try launching and see what happens, you would also need it’s pip dependencies working. You’ll also need to build NVcaffe for ARM, but try using NVcaffe branch NVcaffe-0.15. DIGITS doesn’t maintain official support for ARM and may take significantly longer to complete training, but if starting with a pretrained model (transfer learning) the time may be acceptable.

Are there any steps available to build NVcaffe for ARM such at DIGITS installs and runs, particularly on a Jetson Nano, in this case?

check out this information NVIDIA Container Runtime on Jetson · NVIDIA/nvidia-docker Wiki · GitHub