Wrong architecture of tlt-converter


Managed to train my model, including export.
Now, I want to convert it to run on the Jetson Xavier platform.

Copied the model and the tlt-converter from the docker.

Tried to run on the xavier, and … its the wrong exe :-( :

./tlt-converter: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/l, for GNU/Linux 2.6.32, BuildID[sha1]=a5ad1a9cf44d558ec0477cd0566b544ec8c3cbb4, not stripped

Any ideas where is the correct converter exe ?

Thanks, Best

Hi Danny,

If you don’t mind me asking…

Where did you store your data sets on the docker container? Also how did you manage to migrate them from your personal home computer to the docker container? Is there a command for that?

Lastly, how do we go about setting up the spec.cfg file and editing it? They show us an example one but I am not sure how I can actually go about making one myself.

Thank you for your responses :)



Data - use the shared folders (one that is common to the host and docker, given as param to the docker run cmd).
Setting cfg - use the notebook they supply, modify their cfg files (i think u need to modify 5 of them).

Note also - this is “half baked” at best - the output of this entire exercise is something (hopefully) digestible by DeepStream, BUT JetPack4.2 is currently not supporting DeepStream properly (see other q I asked in the DS forum).

Thanks, Best

Great! Thanks for the information. Everything makes sense.

If my shared folder is locked to any use outside of the container, how would one transfer the data outside of the container into that shared folder?

sudo ?

Got it working. Thanks Danny!

Hey Danny,

Hope you are well. I am at the same stage as you now on running it on Xavier. I have loaded 4.1 on the board to see if it will run there since Deepstream doesnt support 4.2

What commands are you using for the following?
1.Locate the tlt-converter inside your inference environment and add its parent
directory to the system path.
2. Run the converter: tlt-converter …

I want to compare them to what I am attempting to do to make sure I have got all the fields right.

Thanks a bunch!



If I understand correctly, you need to take it from the docker, and copy it to the jetson.

BUT it will not run, its the wrong architecture (x86, while jetson is ARM :-( )

This is the topic of the original thread.

Anyone - any idea what am I doing wrong ?

Thanks, Best

Hi Danny,

Unfortunately, our current release doesn’t support Jetson (only DeepStream on x86). We will add support for Jetson in next release.