JetPack 4.2 — L4T R32.1 release for Jetson AGX Xavier, Jetson TX2, and Jetson Nano

How to install tensorflow with this version of JetPack? None of the old methods work now.

Re: flashing problems. It seemed to me that with the R32.1 release, using JetPack,
it wasn’t possible to ssh to the device, and the usb ‘gadget’ devices (net, serial,
memory) weren’t enabled until after I connected a monitor and keyboard and went through
the Ubuntu setup procedure. Could this be impacting the install procedure?

My solution (described above) was to go through the initial setup once with mouse,
keyboard, and monitor and create some accounts, then save the APP image and use
that as system.img for further booting. Once this was working I didn’t spend any
time investigating alternate procedures, so I’m not 100% sure this is the case.

Hi Victor, JetPack 4.2 has migrated to Ubuntu 18.04 for the Jetson’s OS version.

Hi Dan, have you tried the TensorFlow installer from this thread for JetPack 4.2 and Python 3.6?

Cool I will install that. Wasn’t up there when I tried before I figured it was 42 instead of 33 in the link name :) I just went back to the last jetpack so I could get it done in time. Its at the Science museum now as an AI demo. Running a ZED so I’m getting object identification and ranging via the stereo camera.

I just installed TensorFlow last week using the string in the thread dusty_nv mentioned. FYI, dusty_nv’s github is a gold mine.

I’ve been thinking, isn’t referring to the tensor flow package, or anything else, by its associated jetpack version (ie pip3 install …/jp/v42 or pip3 install …/jp33) a bit of a misnomer? My understanding is that the jetson board doesnt run Jetpack; it runs L4T (ie L4T v32.1 gets installed with jp v4.2). Shouldn’t these packages be refered to their L4T version?

I am new to OpenCV and to the Jetson TX2. I noticed that when I tried to run a python OpenCV script, the GPU was not being used at all. Does the version of OpenCV installed with Jetpack 4.2 support CUDA?

I have seen tutorials on how to build newer versions of OpenCV with CUDA support for the TX2, however they are for older version of the OS, and I am not sure if they are still recommended.

Also, this may be a dumb question, but are newer version of OpenCV compatible with the TX2? More specifically 4.1.0, or should I stick with the older ones?

Hi Alvaro, the default version of OpenCV installed doesn’t have CUDA enabled, so you would need to re-build it with CUDA compilation flag turned on.

Yes, you should be able to use newer versions of OpenCV with Jetson too.

gstreamer v4l2src doesn’t seem to exist. Not sure how our video capture application
can work without this.

Followup: This has been solved see this thread for details

https://devtalk.nvidia.com/default/topic/1050903/jetson-tx2/problem-porting-csi-camera-driver-from-4-4-kernel-to-4-9-kernel/post/5334165/#5334165

(gst-inspect-1.0 for some reason decides to open the v4l2 input devices. somewhat strange)

The “TensorFlow for JetPack” pip wheels are created for a particular JetPack release, and verified to work with the JetPack component versions (i.e., L4T, cuDNN, and TensorRT) included in that release.

https://developer.nvidia.com/embedded/faq#what-is-jetpack
https://developer.nvidia.com/embedded/downloads#?search=TensorFlow

@prlawrence, Oh I see. L4T is the OS, which could be downloaded and flashed stand alone. Where as, jetpack is a specific software package including L4T, cuDNN, TensorRT, etc. And the TensorFlow pip wheels are designed and tested to fit into that software environment. Got it, thanks!

Hello Sir,

Can anyone please tell me which python version comes with the jetpack 4.2 release fot the jetson TX2 board.

Thank you

Hello Sir,
Can you please tell me which version of python comes with the jetpack 4.2 . and with which version of python, the opencv works.

thanks

Hello Sir,
Can you please tell me which version of python comes with the jetpack 4.2 . and with which version of python, the opencv works.

thanks

Hello Sir,
Can you please tell the detailed procedure to install the opencv with cuda enabled.

sdkManager for JetPack 4.2 running on Desktop Ubuntu 16.04, target is Tx2. I get this error during download step

"13:15:46 INFO : CUDA on Host : start to download https://developer.nvidia.com/assets/embedded/secure/tools/files/jetpack-sdks/jetpack-4.2/JETPACK_42_b158/ubuntu1604/cuda-repo-ubuntu1604-10-0-local-10.0.166-410.62_1.0-1_amd64.deb to /home/krell/Downloads/nvidia/sdkm_downloads/cuda-repo-ubuntu1604-10-0-local-10.0.166-410.62_1.0-1_amd64.deb
13:15:51 WARNING : CUDA on Host : retry to connect https://developer.nvidia.com/assets/embedded/secure/tools/files/jetpack-sdks/jetpack-4.2/JETPACK_42_b158/ubuntu1604/cuda-repo-ubuntu1604-10-0-local-10.0.166-410.62_1.0-1_amd64.deb caused by Error: DevZone get download url failed: Error: unable to get local issuer certificate.
13:15:56 WARNING : CUDA on Host : retry to connect https://developer.nvidia.com/assets/embedded/secure/tools/files/jetpack-sdks/jetpack-4.2/JETPACK_42_b158/ubuntu1604/cuda-repo-ubuntu1604-10-0-local-10.0.166-410.62_1.0-1_amd64.deb caused by Error: The .mtd file is corrupt. Start a new download.
13:16:01 WARNING : CUDA on Host : retry to connect https://developer.nvidia.com/assets/embedded/secure/tools/files/jetpack-sdks/jetpack-4.2/JETPACK_42_b158/ubuntu1604/cuda-repo-ubuntu1604-10-0-local-10.0.166-410.62_1.0-1_amd64.deb caused by Error: The .mtd file is corrupt. Start a new download.
13:16:01 ERROR : CUDA on Host : download error: https://developer.nvidia.com/assets/embedded/secure/tools/files/jetpack-sdks/jetpack-4.2/JETPACK_42_b158/ubuntu1604/cuda-repo-ubuntu1604-10-0-local-10.0.166-410.62_1.0-1_amd64.deb failed with undefined

Edit to add: This appears to be how our corporate firewalls handle certificates. I am leaving comment in place to illustrate how tools need to account for modern security firewall boundaries at corporate level.

Why can’t I run lspci with this new version?

“lspci” works on all Jetsons and releases so far as I know. If no device is detected you might find the power bus disabled and no output, but the result is still accurate (other than not showing the root complex due to being shut down). To get max verbosity you do need sudo. Is lspci returning nothing, or is it actually giving an error message? Returning “nothing” would be correct for no devices plus power off to PCIe as an energy savings measure.

I just installed the JetPack 4.2 SDK on a TX2 device to test the new features.

We were using JetPack 3.3 previously with TX1 devices and more specific the Argus library for camera control.

Now I cannot find the tegra_multimedia_api folder in order to build the argus library.

Anyone knows where it is ?

Hi pavlos, you should be able to find the updated location at:

/usr/src/tegra_multimedia_api

do you have a guide or instructions how to upgrade TX2 devices in the field running jetpack 3.3 to jetpack 4.2 ?