ml6team
To make it work with DS 6.2
Maybe this should be comment rather than an issue? (Sorry for 1 more issue, but I hope it's useful)
To make the Dockefile work with DS 6.2, there are 3 important changes that are needed
Obviously, the first line should change from:
FROM nvcr.io/nvidia/deepstream:6.1-devel
to
FROM nvcr.io/nvidia/deepstream:6.2-devel
But there are 2 more things:
-
Add this line after line 7:
ENV CUDA_MODULE_LOADING=LAZYThat's new in 6.2 and I presume it will be ignored in 6.1 and before -
Add this line between current line 15 and 16:
RUN bash /opt/nvidia/deepstream/deepstream/user_additional_install.shAs that will install the extra libraries taht are no longer part of the default package since 6.2
With each update of Deepstream, thorough testing is needed as NVIDIA tends to introduce unwanted breaking changes. If you've tested the codebase with the changes above, feel free to submit a PR. I'd be happy to help update the codebase to version 6.2.
Hi Joxis,
Yes, with those changes, it worked for me in 6.2. (I wish I could say it more certain, but still learning)
Item 2 would break backwards compatiblity. I thought the line below would work, but it is not allowed:
RUN bash test -f /opt/nvidia/deepstream/deepstream/user_additional_install.sh && /opt/nvidia/deepstream/user_additional_install.sh
The only solution I found was here: Conditional Logic In Dockerfile The script would be the same for 6.1 and 6.2 through an extra parameter, but not sure if you want to go that may, making the script more complicated.
Hopefully you have more experience and know an easy solution?
Do you have a repo fork somewhere with code to get started? If so, I could take a look
Hi Joxis,
I'm afraid not. I was just doing some edits locally to make it work and understand how it works. But all the changes to 6.2 are mentioned in the first post. Just the additional libraries are the issue.
(Feel free to close the tickets if it's not worth it. I'll take no offense)
I think an update is useful but unfortunately I currently do not have the capacity to work on this myself. If anyone submits a PR, I'd be happy to review and incorporate it.
How to use this repository with the yolov5 model as well as the custom yolov5 model?
